diff --git a/ckpts/4b_stock/args.json b/ckpts/4b_stock/args.json
deleted file mode 100644
index 8c1cdfe510a40fc54ad5bac2741d3f5659de7620..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/args.json
+++ /dev/null
@@ -1,376 +0,0 @@
-{
-  "output_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_qwen3_5vit_stock_baseline_a800_10pct/v0-20260530-001435",
-  "per_device_train_batch_size": 1,
-  "num_train_epochs": 3.0,
-  "max_steps": 505,
-  "learning_rate": 1e-05,
-  "lr_scheduler_type": "cosine",
-  "lr_scheduler_kwargs": null,
-  "warmup_steps": 0,
-  "optim": "adamw_torch_fused",
-  "optim_args": null,
-  "weight_decay": 0.1,
-  "adam_beta1": 0.9,
-  "adam_beta2": 0.95,
-  "adam_epsilon": 1e-08,
-  "optim_target_modules": null,
-  "gradient_accumulation_steps": 8,
-  "average_tokens_across_devices": true,
-  "max_grad_norm": 1.0,
-  "label_smoothing_factor": 0.0,
-  "bf16": true,
-  "fp16": false,
-  "bf16_full_eval": false,
-  "fp16_full_eval": false,
-  "tf32": null,
-  "gradient_checkpointing": true,
-  "gradient_checkpointing_kwargs": "{\"use_reentrant\": false}",
-  "torch_compile": false,
-  "torch_compile_backend": null,
-  "torch_compile_mode": null,
-  "use_liger_kernel": false,
-  "liger_kernel_config": null,
-  "use_cache": false,
-  "neftune_noise_alpha": null,
-  "torch_empty_cache_steps": null,
-  "auto_find_batch_size": false,
-  "logging_strategy": "steps",
-  "logging_steps": 1,
-  "logging_first_step": true,
-  "log_on_each_node": true,
-  "logging_nan_inf_filter": true,
-  "include_num_input_tokens_seen": false,
-  "log_level": "passive",
-  "log_level_replica": "warning",
-  "disable_tqdm": null,
-  "report_to": [
-    "none"
-  ],
-  "run_name": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_qwen3_5vit_stock_baseline_a800_10pct/v0-20260530-001435",
-  "project": "huggingface",
-  "trackio_space_id": "trackio",
-  "eval_strategy": "no",
-  "eval_steps": 505.0,
-  "eval_delay": 0,
-  "per_device_eval_batch_size": 1,
-  "prediction_loss_only": false,
-  "eval_on_start": false,
-  "eval_do_concat_batches": true,
-  "eval_use_gather_object": false,
-  "eval_accumulation_steps": null,
-  "include_for_metrics": [],
-  "batch_eval_metrics": false,
-  "save_only_model": false,
-  "save_strategy": "steps",
-  "save_steps": 505.0,
-  "save_on_each_node": false,
-  "save_total_limit": 1,
-  "enable_jit_checkpoint": false,
-  "push_to_hub": false,
-  "hub_token": null,
-  "hub_private_repo": null,
-  "hub_model_id": null,
-  "hub_strategy": "every_save",
-  "hub_always_push": false,
-  "hub_revision": null,
-  "load_best_model_at_end": false,
-  "metric_for_best_model": "loss",
-  "greater_is_better": false,
-  "ignore_data_skip": false,
-  "restore_callback_states_from_checkpoint": false,
-  "full_determinism": false,
-  "seed": 42,
-  "data_seed": 42,
-  "use_cpu": false,
-  "accelerator_config": "{\"dispatch_batches\": false}",
-  "parallelism_config": null,
-  "dataloader_drop_last": false,
-  "dataloader_num_workers": 6,
-  "dataloader_pin_memory": true,
-  "dataloader_persistent_workers": true,
-  "dataloader_prefetch_factor": 4,
-  "remove_unused_columns": true,
-  "label_names": null,
-  "train_sampling_strategy": "random",
-  "length_column_name": "length",
-  "ddp_find_unused_parameters": null,
-  "ddp_bucket_cap_mb": null,
-  "ddp_broadcast_buffers": null,
-  "ddp_backend": null,
-  "ddp_timeout": 7200,
-  "fsdp": [],
-  "fsdp_config": null,
-  "deepspeed": {
-    "fp16": {
-      "enabled": "auto",
-      "loss_scale": 0,
-      "loss_scale_window": 1000,
-      "initial_scale_power": 16,
-      "hysteresis": 2,
-      "min_loss_scale": 1
-    },
-    "bf16": {
-      "enabled": "auto"
-    },
-    "zero_optimization": {
-      "stage": 2,
-      "offload_optimizer": {
-        "device": "none",
-        "pin_memory": true
-      },
-      "allgather_partitions": true,
-      "allgather_bucket_size": 200000000.0,
-      "overlap_comm": false,
-      "reduce_scatter": true,
-      "reduce_bucket_size": 200000000.0,
-      "contiguous_gradients": true
-    },
-    "gradient_accumulation_steps": "auto",
-    "gradient_clipping": "auto",
-    "steps_per_print": 2000,
-    "train_batch_size": "auto",
-    "train_micro_batch_size_per_gpu": "auto",
-    "wall_clock_breakdown": false
-  },
-  "debug": null,
-  "skip_memory_metrics": true,
-  "do_train": false,
-  "do_eval": false,
-  "do_predict": false,
-  "resume_from_checkpoint": null,
-  "warmup_ratio": 0.05,
-  "logging_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_qwen3_5vit_stock_baseline_a800_10pct/v0-20260530-001435/runs",
-  "local_rank": 0,
-  "sortish_sampler": false,
-  "predict_with_generate": false,
-  "generation_max_length": null,
-  "generation_num_beams": null,
-  "generation_config": null,
-  "tuner_backend": "peft",
-  "vit_gradient_checkpointing": true,
-  "router_aux_loss_coef": 0.0,
-  "enable_dft_loss": false,
-  "enable_channel_loss": false,
-  "safe_serialization": true,
-  "max_shard_size": "5GB",
-  "check_model": true,
-  "acc_strategy": "token",
-  "train_dataloader_shuffle": true,
-  "group_by_length": false,
-  "max_epochs": null,
-  "aligner_lr": null,
-  "vit_lr": 1e-06,
-  "use_logits_to_keep": null,
-  "ds3_gather_for_generation": true,
-  "resume_only_model": false,
-  "optimizer": null,
-  "loss_type": null,
-  "eval_metric": null,
-  "callbacks": [],
-  "early_stop_interval": null,
-  "eval_use_evalscope": false,
-  "eval_dataset": [],
-  "eval_dataset_args": null,
-  "eval_limit": null,
-  "eval_generation_config": null,
-  "extra_eval_args": null,
-  "tuner_type": "full",
-  "use_galore": false,
-  "galore_target_modules": null,
-  "galore_rank": 128,
-  "galore_update_proj_gap": 50,
-  "galore_scale": 1.0,
-  "galore_proj_type": "std",
-  "galore_optim_per_parameter": false,
-  "galore_with_embedding": false,
-  "galore_quantization": false,
-  "galore_proj_quant": false,
-  "galore_proj_bits": 4,
-  "galore_proj_group_size": 256,
-  "galore_cos_threshold": 0.4,
-  "galore_gamma_proj": 2,
-  "galore_queue_size": 5,
-  "lisa_activated_layers": 0,
-  "lisa_step_interval": 20,
-  "use_flash_ckpt": false,
-  "use_ray": false,
-  "ray_exp_name": null,
-  "device_groups": null,
-  "model": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_qwen3_5vit_stock_baseline_a800/v0-20260529-170602/checkpoint-2181",
-  "model_type": "llava_qwen3_5vit_qwen3",
-  "model_revision": null,
-  "task_type": "causal_lm",
-  "torch_dtype": "bfloat16",
-  "attn_impl": "flash_attn",
-  "experts_impl": null,
-  "new_special_tokens": [],
-  "num_labels": null,
-  "problem_type": null,
-  "rope_scaling": null,
-  "device_map": null,
-  "max_memory": {},
-  "max_model_len": null,
-  "local_repo_path": null,
-  "init_strategy": null,
-  "template": "llava_qwen3_5vit_qwen3",
-  "system": null,
-  "max_length": 16384,
-  "truncation_strategy": "delete",
-  "max_pixels": null,
-  "agent_template": null,
-  "norm_bbox": null,
-  "use_chat_template": true,
-  "padding_side": "right",
-  "padding_free": true,
-  "loss_scale": "default",
-  "sequence_parallel_size": 1,
-  "template_backend": "swift",
-  "response_prefix": null,
-  "enable_thinking": null,
-  "add_non_thinking_prefix": true,
-  "dataset": [],
-  "val_dataset": [],
-  "cached_dataset": [
-    "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/image_10pct/train",
-    "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/video_10pct/train"
-  ],
-  "cached_val_dataset": [],
-  "split_dataset_ratio": 0.0,
-  "dataset_num_proc": 16,
-  "load_from_cache_file": false,
-  "dataset_shuffle": true,
-  "val_dataset_shuffle": false,
-  "streaming": false,
-  "interleave_prob": null,
-  "stopping_strategy": "first_exhausted",
-  "shuffle_buffer_size": 1000,
-  "download_mode": "reuse_dataset_if_exists",
-  "columns": {},
-  "strict": false,
-  "model_name": null,
-  "model_author": null,
-  "custom_dataset_info": [],
-  "quant_method": null,
-  "quant_bits": null,
-  "hqq_axis": null,
-  "bnb_4bit_compute_dtype": "bfloat16",
-  "bnb_4bit_quant_type": "nf4",
-  "bnb_4bit_use_double_quant": true,
-  "bnb_4bit_quant_storage": null,
-  "max_new_tokens": 64,
-  "temperature": 0.0,
-  "top_k": null,
-  "top_p": null,
-  "repetition_penalty": null,
-  "num_beams": 1,
-  "stream": false,
-  "stop_words": [],
-  "logprobs": false,
-  "top_logprobs": null,
-  "structured_outputs_regex": null,
-  "adapters": [],
-  "external_plugins": [
-    "ms-swift/video_mllm/model_plugin_qwen3_5vit.py",
-    "ms-swift/video_mllm/dataset_plugin.py"
-  ],
-  "custom_register_path": [],
-  "model_kwargs": {},
-  "load_args": false,
-  "load_data_args": false,
-  "packing": true,
-  "packing_length": 16384,
-  "packing_num_proc": 1,
-  "lazy_tokenize": false,
-  "use_hf": true,
-  "ignore_args_error": false,
-  "use_swift_lora": false,
-  "freeze_parameters": [],
-  "freeze_parameters_regex": null,
-  "freeze_parameters_ratio": 0.0,
-  "trainable_parameters": [
-    "model.multi_modal_projector"
-  ],
-  "trainable_parameters_regex": null,
-  "freeze_llm": false,
-  "freeze_vit": false,
-  "freeze_aligner": false,
-  "target_modules": [
-    "all-linear"
-  ],
-  "target_regex": null,
-  "target_parameters": null,
-  "modules_to_save": [],
-  "lora_rank": 8,
-  "lora_alpha": 32,
-  "lora_dropout": 0.05,
-  "lora_bias": "none",
-  "lora_dtype": null,
-  "lorap_lr_ratio": null,
-  "use_rslora": false,
-  "use_dora": false,
-  "lora_ga_batch_size": 2,
-  "lora_ga_iters": 2,
-  "lora_ga_max_length": 1024,
-  "lora_ga_direction": "ArB2r",
-  "lora_ga_scale": "stable",
-  "lora_ga_stable_gamma": 16,
-  "init_weights": true,
-  "fourier_n_frequency": 2000,
-  "fourier_scaling": 300.0,
-  "boft_block_size": 4,
-  "boft_block_num": 0,
-  "boft_n_butterfly_factor": 1,
-  "boft_dropout": 0.0,
-  "vera_rank": 256,
-  "vera_projection_prng_key": 0,
-  "vera_dropout": 0.0,
-  "vera_d_initial": 0.1,
-  "adapter_act": "gelu",
-  "adapter_length": 128,
-  "adalora_target_r": 8,
-  "adalora_init_r": 12,
-  "adalora_tinit": 0,
-  "adalora_tfinal": 0,
-  "adalora_deltaT": 1,
-  "adalora_beta1": 0.85,
-  "adalora_beta2": 0.85,
-  "adalora_orth_reg_weight": 0.5,
-  "llamapro_num_new_blocks": 4,
-  "llamapro_num_groups": null,
-  "reft_layer_key": null,
-  "reft_layers": null,
-  "reft_rank": 4,
-  "reft_intervention_type": "LoreftIntervention",
-  "reft_args": null,
-  "swanlab_token": null,
-  "swanlab_project": "ms-swift",
-  "swanlab_workspace": null,
-  "swanlab_exp_name": null,
-  "swanlab_notification_method": null,
-  "swanlab_webhook_url": null,
-  "swanlab_secret": null,
-  "swanlab_sender_email": null,
-  "swanlab_receiver_email": null,
-  "swanlab_smtp_server": null,
-  "swanlab_smtp_port": null,
-  "swanlab_email_language": "zh",
-  "swanlab_mode": "cloud",
-  "add_version": true,
-  "create_checkpoint_symlink": false,
-  "zero_hpz_partition_size": null,
-  "deepspeed_autotp_size": null,
-  "swift_version": "4.1.1",
-  "ckpt_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_qwen3_5vit_stock_baseline_a800/v0-20260529-170602/checkpoint-2181",
-  "rank": 0,
-  "global_world_size": 8,
-  "local_world_size": 8,
-  "model_suffix": "checkpoint-2181",
-  "model_info": "ModelInfo(model_type='llava_qwen3_5vit_qwen3', model_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_qwen3_5vit_stock_baseline_a800/v0-20260529-170602/checkpoint-2181', torch_dtype=torch.bfloat16, max_model_len=262144, quant_method=None, quant_bits=None, rope_scaling=None, is_moe_model=False, is_multimodal=True, config=None, task_type='causal_lm', num_labels=None)",
-  "model_meta": "ModelMeta(model_type='llava_qwen3_5vit_qwen3', model_groups=[ModelGroup(models=[Model(ms_model_id='Qwen/Qwen3-0.6B', hf_model_id='Qwen/Qwen3-0.6B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-1.7B', hf_model_id='Qwen/Qwen3-1.7B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-4B', hf_model_id='Qwen/Qwen3-4B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-8B', hf_model_id='Qwen/Qwen3-8B', model_path=None, ms_revision=None, hf_revision=None)], template=None, ignore_patterns=None, requires=None, tags=[])], loader=<class 'model_plugin_qwen3_5vit.Qwen3_5ViTQwen3Loader'>, template='llava_qwen3_5vit_qwen3', model_arch=MultiModelKeys(arch_name='llava_hf', embedding=None, module_list=None, lm_head=None, q_proj=None, k_proj=None, v_proj=None, o_proj=None, attention=None, mlp=None, down_proj=None, qkv_proj=None, qk_proj=None, qa_proj=None, qb_proj=None, kv_proj=None, kva_proj=None, kvb_proj=None, language_model=['model.language_model', 'lm_head'], aligner=['model.multi_modal_projector'], vision_tower=['model.vision_tower'], generator=[]), mcore_model_type=None, architectures=['LlavaQwen3_5ViTForConditionalGeneration'], additional_saved_files=[], torch_dtype=None, is_multimodal=True, is_reward=False, task_type=None, ignore_patterns=None, requires=[], tags=['vision', 'video'])",
-  "model_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_qwen3_5vit_stock_baseline_a800/v0-20260529-170602/checkpoint-2181",
-  "template_meta": "QwenTemplateMeta(template_type='llava_qwen3_5vit_qwen3', prefix=[], prompt=['<|im_start|>user\\n{{QUERY}}<|im_end|>\\n<|im_start|>assistant\\n'], chat_sep=['<|im_end|>\\n'], suffix=['<|im_end|>\\n'], template_cls=<class 'model_plugin_qwen3_5vit.Qwen3_5ViTLlavaTemplate'>, system_prefix=['<|im_start|>system\\n{{SYSTEM}}<|im_end|>\\n'], default_system=None, auto_add_bos=False, stop_words=['<|endoftext|>'], agent_template='hermes', is_thinking=False, thinking_prefix='', non_thinking_prefix='', history_thinking_prefix='')",
-  "_val_dataset_exists": false,
-  "hub": "<class 'swift.hub.hub.HFHub'>",
-  "evaluation_strategy": "steps",
-  "training_args": "Seq2SeqTrainingArguments(output_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_qwen3_5vit_stock_baseline_a800_10pct/v0-20260530-001435', per_device_train_batch_size=1, num_train_epochs=3.0, max_steps=505, learning_rate=1e-05, lr_scheduler_type=<SchedulerType.COSINE: 'cosine'>, lr_scheduler_kwargs=None, warmup_steps=0.05, optim=<OptimizerNames.ADAMW_TORCH_FUSED: 'adamw_torch_fused'>, optim_args=None, weight_decay=0.1, adam_beta1=0.9, adam_beta2=0.95, adam_epsilon=1e-08, optim_target_modules=None, gradient_accumulation_steps=8, average_tokens_across_devices=None, max_grad_norm=1.0, label_smoothing_factor=0.0, bf16=True, fp16=False, bf16_full_eval=False, fp16_full_eval=False, tf32=None, gradient_checkpointing=True, gradient_checkpointing_kwargs={'use_reentrant': False}, torch_compile=False, torch_compile_backend=None, torch_compile_mode=None, use_liger_kernel=False, liger_kernel_config=None, use_cache=False, neftune_noise_alpha=None, torch_empty_cache_steps=None, auto_find_batch_size=False, logging_strategy=<IntervalStrategy.STEPS: 'steps'>, logging_steps=1, logging_first_step=True, log_on_each_node=True, logging_nan_inf_filter=True, include_num_input_tokens_seen=None, log_level='passive', log_level_replica='warning', disable_tqdm=False, report_to=[], run_name='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_qwen3_5vit_stock_baseline_a800_10pct/v0-20260530-001435', project='huggingface', trackio_space_id='trackio', eval_strategy=<IntervalStrategy.NO: 'no'>, eval_steps=505.0, eval_delay=0, per_device_eval_batch_size=1, prediction_loss_only=False, eval_on_start=False, eval_do_concat_batches=True, eval_use_gather_object=False, eval_accumulation_steps=None, include_for_metrics=[], batch_eval_metrics=False, save_only_model=False, save_strategy=<SaveStrategy.STEPS: 'steps'>, save_steps=505, save_on_each_node=False, save_total_limit=1, enable_jit_checkpoint=False, push_to_hub=False, hub_token=None, hub_private_repo=None, hub_model_id=None, hub_strategy=<HubStrategy.EVERY_SAVE: 'every_save'>, hub_always_push=False, hub_revision=None, load_best_model_at_end=False, metric_for_best_model='loss', greater_is_better=False, ignore_data_skip=False, restore_callback_states_from_checkpoint=False, full_determinism=False, seed=42, data_seed=42, use_cpu=False, accelerator_config=AcceleratorConfig(split_batches=False, dispatch_batches=False, even_batches=True, use_seedable_sampler=True, non_blocking=False, gradient_accumulation_kwargs=None, use_configured_state=False), parallelism_config=None, dataloader_drop_last=False, dataloader_num_workers=6, dataloader_pin_memory=True, dataloader_persistent_workers=True, dataloader_prefetch_factor=4, remove_unused_columns=False, label_names=None, train_sampling_strategy='random', length_column_name='length', ddp_find_unused_parameters=None, ddp_bucket_cap_mb=None, ddp_broadcast_buffers=None, ddp_backend=None, ddp_timeout=7200, fsdp=[], fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}, deepspeed={'fp16': {'enabled': 'auto', 'loss_scale': 0, 'loss_scale_window': 1000, 'initial_scale_power': 16, 'hysteresis': 2, 'min_loss_scale': 1}, 'bf16': {'enabled': 'auto'}, 'zero_optimization': {'stage': 2, 'offload_optimizer': {'device': 'none', 'pin_memory': True}, 'allgather_partitions': True, 'allgather_bucket_size': 200000000.0, 'overlap_comm': False, 'reduce_scatter': True, 'reduce_bucket_size': 200000000.0, 'contiguous_gradients': True}, 'gradient_accumulation_steps': 'auto', 'gradient_clipping': 'auto', 'steps_per_print': 2000, 'train_batch_size': 'auto', 'train_micro_batch_size_per_gpu': 'auto', 'wall_clock_breakdown': False}, debug=[], skip_memory_metrics=True, do_train=False, do_eval=False, do_predict=False, resume_from_checkpoint=None, warmup_ratio=0.05, logging_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_qwen3_5vit_stock_baseline_a800_10pct/v0-20260530-001435/runs', local_rank=0, sortish_sampler=False, predict_with_generate=False, generation_max_length=None, generation_num_beams=None, generation_config=None, tuner_backend='peft', vit_gradient_checkpointing=True, router_aux_loss_coef=0.0, enable_dft_loss=False, enable_channel_loss=False, safe_serialization=True, max_shard_size='5GB', check_model=True, acc_strategy='token', train_dataloader_shuffle=True, group_by_length=False, max_epochs=None, aligner_lr=None, vit_lr=1e-06, use_logits_to_keep=None, ds3_gather_for_generation=True, resume_only_model=False, optimizer='multimodal', loss_type=None, eval_metric=None, callbacks=[], early_stop_interval=None, eval_use_evalscope=False, eval_dataset=[], eval_dataset_args=None, eval_limit=None, eval_generation_config=None, extra_eval_args=None, tuner_type='full', use_galore=False, galore_target_modules=None, galore_rank=128, galore_update_proj_gap=50, galore_scale=1.0, galore_proj_type='std', galore_optim_per_parameter=False, galore_with_embedding=False, galore_quantization=False, galore_proj_quant=False, galore_proj_bits=4, galore_proj_group_size=256, galore_cos_threshold=0.4, galore_gamma_proj=2, galore_queue_size=5, lisa_activated_layers=0, lisa_step_interval=20, use_flash_ckpt=False)"
-}
\ No newline at end of file
diff --git a/ckpts/4b_stock/chat_template.jinja b/ckpts/4b_stock/chat_template.jinja
deleted file mode 100644
index 70adff8a08fb31e0636f618564838d4bf3c05286..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/chat_template.jinja
+++ /dev/null
@@ -1,61 +0,0 @@
-{%- if tools %}
-    {{- '<|im_start|>system\n' }}
-    {%- if messages[0].role == 'system' %}
-        {{- messages[0].content + '\n\n' }}
-    {%- endif %}
-    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
-    {%- for tool in tools %}
-        {{- "\n" }}
-        {{- tool | tojson }}
-    {%- endfor %}
-    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
-{%- else %}
-    {%- if messages[0].role == 'system' %}
-        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
-    {%- endif %}
-{%- endif %}
-{%- for message in messages %}
-    {%- if message.content is string %}
-        {%- set content = message.content %}
-    {%- else %}
-        {%- set content = '' %}
-    {%- endif %}
-    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
-        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
-    {%- elif message.role == "assistant" %}
-        {{- '<|im_start|>' + message.role + '\n' + content }}
-        {%- if message.tool_calls %}
-            {%- for tool_call in message.tool_calls %}
-                {%- if (loop.first and content) or (not loop.first) %}
-                    {{- '\n' }}
-                {%- endif %}
-                {%- if tool_call.function %}
-                    {%- set tool_call = tool_call.function %}
-                {%- endif %}
-                {{- '<tool_call>\n{"name": "' }}
-                {{- tool_call.name }}
-                {{- '", "arguments": ' }}
-                {%- if tool_call.arguments is string %}
-                    {{- tool_call.arguments }}
-                {%- else %}
-                    {{- tool_call.arguments | tojson }}
-                {%- endif %}
-                {{- '}\n</tool_call>' }}
-            {%- endfor %}
-        {%- endif %}
-        {{- '<|im_end|>\n' }}
-    {%- elif message.role == "tool" %}
-        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
-            {{- '<|im_start|>user' }}
-        {%- endif %}
-        {{- '\n<tool_response>\n' }}
-        {{- content }}
-        {{- '\n</tool_response>' }}
-        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
-            {{- '<|im_end|>\n' }}
-        {%- endif %}
-    {%- endif %}
-{%- endfor %}
-{%- if add_generation_prompt %}
-    {{- '<|im_start|>assistant\n' }}
-{%- endif %}
\ No newline at end of file
diff --git a/ckpts/4b_stock/chat_template.json b/ckpts/4b_stock/chat_template.json
deleted file mode 100644
index 68724c68b1612e9ab2e31f84e3279099ad656ff6..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/chat_template.json
+++ /dev/null
@@ -1,3 +0,0 @@
-{
-  "chat_template": "{% for message in messages %}{{'<|im_start|>' + message['role'] + ' '}}{# Render all images first #}{% for content in message['content'] | selectattr('type', 'equalto', 'image') %}{{ '<image>' }}{% endfor %}{# Render all video then #}{% for content in message['content'] | selectattr('type', 'equalto', 'video') %}{{ '<video>' }}{% endfor %}{# Render all text next #}{% if message['role'] != 'assistant' %}{% for content in message['content'] | selectattr('type', 'equalto', 'text') %}{{ '\n' + content['text'] }}{% endfor %}{% else %}{% for content in message['content'] | selectattr('type', 'equalto', 'text') %}{% generation %}{{ '\n' + content['text'] }}{% endgeneration %}{% endfor %}{% endif %}{{'<|im_end|>'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"
-}
diff --git a/ckpts/4b_stock/config.json b/ckpts/4b_stock/config.json
deleted file mode 100644
index ce830f6986f152f6aeb43595f9a789a81c62b646..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/config.json
+++ /dev/null
@@ -1,262 +0,0 @@
-{
-  "architectures": [
-    "LlavaQwen3_5ViTForConditionalGeneration"
-  ],
-  "bos_token_id": null,
-  "dtype": "bfloat16",
-  "eos_token_id": 151645,
-  "hidden_size": 2560,
-  "image_grid_pinpoints": [
-    [
-      384,
-      384
-    ],
-    [
-      384,
-      768
-    ],
-    [
-      384,
-      1152
-    ],
-    [
-      384,
-      1536
-    ],
-    [
-      384,
-      1920
-    ],
-    [
-      384,
-      2304
-    ],
-    [
-      768,
-      384
-    ],
-    [
-      768,
-      768
-    ],
-    [
-      768,
-      1152
-    ],
-    [
-      768,
-      1536
-    ],
-    [
-      768,
-      1920
-    ],
-    [
-      768,
-      2304
-    ],
-    [
-      1152,
-      384
-    ],
-    [
-      1152,
-      768
-    ],
-    [
-      1152,
-      1152
-    ],
-    [
-      1152,
-      1536
-    ],
-    [
-      1152,
-      1920
-    ],
-    [
-      1152,
-      2304
-    ],
-    [
-      1536,
-      384
-    ],
-    [
-      1536,
-      768
-    ],
-    [
-      1536,
-      1152
-    ],
-    [
-      1536,
-      1536
-    ],
-    [
-      1536,
-      1920
-    ],
-    [
-      1536,
-      2304
-    ],
-    [
-      1920,
-      384
-    ],
-    [
-      1920,
-      768
-    ],
-    [
-      1920,
-      1152
-    ],
-    [
-      1920,
-      1536
-    ],
-    [
-      1920,
-      1920
-    ],
-    [
-      1920,
-      2304
-    ],
-    [
-      2304,
-      384
-    ],
-    [
-      2304,
-      768
-    ],
-    [
-      2304,
-      1152
-    ],
-    [
-      2304,
-      1536
-    ],
-    [
-      2304,
-      1920
-    ],
-    [
-      2304,
-      2304
-    ]
-  ],
-  "image_token_index": 151669,
-  "model_type": "llava_qwen3_5vit_qwen3",
-  "multimodal_projector_bias": true,
-  "pad_token_id": 151643,
-  "projector_hidden_act": "gelu",
-  "text_config": {
-    "_name_or_path": "/share/m2v_intern_v3/wangjunjie09/model_cache/huggingface/Qwen/Qwen3-4B-Instruct-2507",
-    "architectures": [
-      "Qwen3ForCausalLM"
-    ],
-    "attention_bias": false,
-    "attention_dropout": 0.0,
-    "bos_token_id": 151643,
-    "dtype": "bfloat16",
-    "eos_token_id": 151645,
-    "head_dim": 128,
-    "hidden_act": "silu",
-    "hidden_size": 2560,
-    "initializer_range": 0.02,
-    "intermediate_size": 9728,
-    "layer_types": [
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention"
-    ],
-    "max_position_embeddings": 262144,
-    "max_window_layers": 36,
-    "model_type": "qwen3",
-    "num_attention_heads": 32,
-    "num_hidden_layers": 36,
-    "num_key_value_heads": 8,
-    "pad_token_id": 151643,
-    "rms_norm_eps": 1e-06,
-    "rope_parameters": {
-      "rope_theta": 5000000,
-      "rope_type": "default"
-    },
-    "sliding_window": null,
-    "tie_word_embeddings": true,
-    "use_cache": false,
-    "use_sliding_window": false,
-    "vocab_size": 151936
-  },
-  "tie_word_embeddings": true,
-  "transformers_version": "5.5.4",
-  "use_cache": false,
-  "video_token_index": 151670,
-  "vision_aspect_ratio": "anyres_max_9",
-  "vision_config": {
-    "deepstack_visual_indexes": [],
-    "depth": 24,
-    "dtype": "bfloat16",
-    "hidden_act": "gelu_pytorch_tanh",
-    "hidden_size": 1024,
-    "image_size": 384,
-    "in_channels": 3,
-    "initializer_range": 0.02,
-    "intermediate_size": 4096,
-    "model_type": "qwen3_5",
-    "num_heads": 16,
-    "num_position_embeddings": 2304,
-    "out_hidden_size": 2048,
-    "patch_size": 16,
-    "spatial_merge_size": 2,
-    "temporal_patch_size": 2
-  },
-  "vision_feature_layer": -1,
-  "vision_feature_select_strategy": "full",
-  "auto_map": {
-    "AutoConfig": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTConfig",
-    "AutoModel": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
-    "AutoModelForCausalLM": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
-    "AutoModelForImageTextToText": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration"
-  }
-}
\ No newline at end of file
diff --git a/ckpts/4b_stock/generation_config.json b/ckpts/4b_stock/generation_config.json
deleted file mode 100644
index 56320d3179df3512d0cb76c240831a10e6f7c6f7..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/generation_config.json
+++ /dev/null
@@ -1,12 +0,0 @@
-{
-  "_from_model_config": true,
-  "bos_token_id": 151643,
-  "eos_token_id": [
-    151645,
-    151643
-  ],
-  "output_attentions": false,
-  "output_hidden_states": false,
-  "transformers_version": "5.5.4",
-  "use_cache": true
-}
diff --git a/ckpts/4b_stock/latest b/ckpts/4b_stock/latest
deleted file mode 100644
index bda446cd4c4d586af0956cf22fb3bd54d8006aec..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/latest
+++ /dev/null
@@ -1 +0,0 @@
-global_step505
\ No newline at end of file
diff --git a/ckpts/4b_stock/model-00001-of-00002.safetensors b/ckpts/4b_stock/model-00001-of-00002.safetensors
deleted file mode 100644
index f05f073b314ab2c3017d3d590c9dec0a21555721..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/model-00001-of-00002.safetensors
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:87bf8644b6e1d76637dd2c54e296952c9906224aa34740c30b757af2623e2a3c
-size 4987507408
diff --git a/ckpts/4b_stock/model-00002-of-00002.safetensors b/ckpts/4b_stock/model-00002-of-00002.safetensors
deleted file mode 100644
index 0052105b800652b6de7ff1e1062be85c7c488e8f..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/model-00002-of-00002.safetensors
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:feb82dce0fbd8cd4e8f27f887a5673c7d519f735e368c5f7fce50663f4317757
-size 4466285248
diff --git a/ckpts/4b_stock/model.safetensors.index.json b/ckpts/4b_stock/model.safetensors.index.json
deleted file mode 100644
index a227833b5ba2c2ae9deeddde8d4fc716babffc8c..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/model.safetensors.index.json
+++ /dev/null
@@ -1,705 +0,0 @@
-{
-  "metadata": {
-    "total_parameters": 4337895424,
-    "total_size": 9453703168
-  },
-  "weight_map": {
-    "lm_head.weight": "model-00001-of-00002.safetensors",
-    "model.image_newline": "model-00001-of-00002.safetensors",
-    "model.language_model.embed_tokens.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.norm.weight": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_1.bias": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_1.weight": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_2.bias": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.final_layernorm.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.final_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.patch_embed.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.patch_embed.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.pos_embed.weight": "model-00002-of-00002.safetensors"
-  }
-}
diff --git a/ckpts/4b_stock/modeling_qwen3_5vit_qwen3.py b/ckpts/4b_stock/modeling_qwen3_5vit_qwen3.py
deleted file mode 100644
index b3f5b907b874422828e6550e4d0e9393d572a96c..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/modeling_qwen3_5vit_qwen3.py
+++ /dev/null
@@ -1,351 +0,0 @@
-"""
-modeling_qwen3_5vit_qwen3.py — Qwen3.5 Vision as SigLIP-compat vision_tower in LlavaOnevision.
-
-设计：与 `modeling_qwen3vlvit_qwen3.py` 严格同构，仅换 vision backbone 源：
-  - Qwen3VLVisionModel → Qwen3_5VisionModel（继承关系：Qwen3_5VisionModel(Qwen3VLVisionModel) 去 DeepStack）
-  - Qwen3VLVisionConfig → Qwen3_5VisionConfig（父类用 AttributeError 哨兵屏蔽 deepstack_visual_indexes）
-
-其余（Adapter 契约翻译、MLP projector + pre_norm、LlavaOnevision 继承 wire class）与
-Qwen3-VL ViT pipeline 完全一致。两条 pipeline 并存意义：DeepStack ablation 天然实验组。
-
-类层级：
-  Qwen3_5ViTBackbone(Qwen3_5VisionModel)      — 去 merger，保持 NaViT 契约
-  Qwen3_5ViTAsSiglipAdapter(nn.Module)        — 持有 Backbone，做 SigLIP ↔ NaViT 契约翻译
-
-三方对比公平性：定 384×384 AnyRes tile + 同款 projector 骨架 + 同款 Qwen3-1.7B LLM。
-"""
-
-import math
-import os
-import sys
-from typing import Optional
-
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-from transformers import (
-    AutoConfig,
-    AutoModel,
-    AutoModelForCausalLM,
-    LlavaOnevisionConfig,
-    LlavaOnevisionForConditionalGeneration,
-    LlavaOnevisionModel,
-    LlavaOnevisionPreTrainedModel,
-    Qwen3Config,
-)
-from transformers.activations import ACT2FN
-from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
-from transformers.models.qwen3_5.configuration_qwen3_5 import Qwen3_5VisionConfig
-from transformers.models.qwen3_5.modeling_qwen3_5 import Qwen3_5VisionModel
-
-# Shared layout-permutation utility lives in declip_qwenvit (single source of
-# truth — same code path runs in declip-training-side qk_cosine reorder).
-# Add VisionEncoder repo root to sys.path so this modeling file is importable
-# even when the package isn't pip-installed (ms-swift integration loads it
-# via dynamic plugin path).
-_REPO_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
-if _REPO_ROOT not in sys.path:
-    sys.path.append(_REPO_ROOT)
-from declip_qwenvit.model.qwen3vit_qk import block_merge_to_row_major_permutation  # noqa: E402
-
-
-class LlavaQwen3_5ViTConfig(LlavaOnevisionConfig):
-    """LlavaOnevisionConfig 子类，vision_config 类型换成 Qwen3_5VisionConfig。
-
-    与 Qwen3-VL ViT 版的差异：
-      - sub_configs.vision_config 用 Qwen3_5VisionConfig
-      - 不再设置 deepstack_visual_indexes（Qwen3_5VisionConfig 用 AttributeError 哨兵屏蔽此字段）
-
-    其余同 LlavaQwen3VLViTConfig（tile_size 默认 384；vision_feature_select_strategy='full'
-    必须 override，Qwen3.5 ViT 无 CLS token）。
-    """
-
-    model_type = "llava_qwen3_5vit_qwen3"
-    sub_configs = {"vision_config": Qwen3_5VisionConfig, "text_config": Qwen3Config}
-
-    def __init__(
-        self,
-        vision_config=None,
-        text_config=None,
-        tile_size: int = 384,
-        **kwargs,
-    ):
-        if isinstance(vision_config, dict):
-            vision_config = Qwen3_5VisionConfig(**vision_config)
-        elif vision_config is None:
-            vision_config = Qwen3_5VisionConfig()
-        # WHY 无 `vision_config.deepstack_visual_indexes = []`（对比 Qwen3-VL ViT 版）：
-        # Qwen3_5VisionConfig 父类用 AttributeError() 哨兵显式屏蔽此字段，设置会报 AttributeError
-        # LlavaOnevision.pack_image_features 用这个作为 tile 像素大小（不是 patch_size）
-        vision_config.image_size = tile_size
-
-        if isinstance(text_config, dict):
-            text_config = Qwen3Config(**text_config)
-        elif text_config is None:
-            text_config = Qwen3Config()
-
-        # 父类默认 select_strategy='default' 会跳首 token (CLS) — Qwen3.5 ViT 无 CLS 必须用 'full'
-        kwargs.setdefault("vision_feature_select_strategy", "full")
-        super().__init__(vision_config=vision_config, text_config=text_config, **kwargs)
-
-
-class Qwen3_5ViTBackbone(Qwen3_5VisionModel):
-    """Qwen3.5 Vision 去除原生 patch_merger 的 backbone 版本（V6 final_layernorm fix (2026-05-16): append final_layernorm）。
-
-    构造时把 merger.norm 的预训练权重抠到 final_layernorm，然后 `del self.merger`
-    释放 ~37M 参数（保留 norm 的 LN 焊到末端做 post_layernorm 角色，丢弃 spatial
-    shuffle + linear_fc1/fc2，那对应 LlavaOV projector 的职责）。
-
-    架构对称（V6 final_layernorm fix (2026-05-16) 修复）：
-      SigLIP2:    encoder → post_layernorm → last_hidden_state → LlavaOV MLP → LLM
-      V6 final_layernorm fix (2026-05-16):     encoder → final_layernorm → last_hidden_state → LlavaOV MLP → LLM
-
-    forward 跑完 transformer blocks 后过 final_layernorm，再返回。下游 LlavaOnevision
-    pack_image_features 的 AnyRes 2×2 pool 接管原 merger 的空间合并职责。
-
-    输入输出契约与父类 Qwen3_5VisionModel 一致（NaViT flat）：
-      forward(hidden_states=[L, patch_dim], grid_thw=[N, 3])
-      → BaseModelOutput(last_hidden_state=[L, hidden_size])
-
-    forward 主体 1:1 对照 `Qwen3_5VisionModel.forward`（已无 deepstack loop，比
-    Qwen3VLVisionModel.forward 更短），仅跳过末尾 `self.merger(x)`，改为 final_layernorm。
-    """
-
-    def __init__(self, config):
-        super().__init__(config)
-        # V6 final_layernorm fix (2026-05-16): extract merger.norm pretrained weights into final_layernorm.
-        # Default init: even if ckpt 阶段没 inject final_layernorm.* (e.g. stock
-        # bootstrap path that filters merger.*), final_layernorm 仍持有 merger.norm
-        # 的预训练值, 不是 random — 这是防止 silent corruption 的兜底.
-        ln_w = self.merger.norm.weight.detach().clone()
-        ln_b = self.merger.norm.bias.detach().clone()
-        del self.merger
-        self.final_layernorm = nn.LayerNorm(config.hidden_size, eps=1e-6)
-        self.final_layernorm.weight.data.copy_(ln_w)
-        self.final_layernorm.bias.data.copy_(ln_b)
-
-    def forward(self, hidden_states, grid_thw, **kwargs):
-        hidden_states = self.patch_embed(hidden_states)
-
-        pos_embeds = self.fast_pos_embed_interpolate(grid_thw)
-        hidden_states = hidden_states + pos_embeds
-
-        rotary_pos_emb = self.rot_pos_emb(grid_thw)
-        seq_len, _ = hidden_states.size()
-        rotary_pos_emb = rotary_pos_emb.reshape(seq_len, -1)
-        emb = torch.cat((rotary_pos_emb, rotary_pos_emb), dim=-1)
-        position_embeddings = (emb.cos(), emb.sin())
-
-        cu_seqlens = torch.repeat_interleave(
-            grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]
-        ).cumsum(dim=0, dtype=torch.int32)
-        cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
-
-        for blk in self.blocks:
-            hidden_states = blk(
-                hidden_states,
-                cu_seqlens=cu_seqlens,
-                position_embeddings=position_embeddings,
-                **kwargs,
-            )
-
-        # V6 final_layernorm fix (2026-05-16): appended final LayerNorm — mirrors SigLIP2's post_layernorm.
-        # Per-token affine; layout-invariant (reorder happens in adapter).
-        hidden_states = self.final_layernorm(hidden_states)
-
-        return BaseModelOutput(last_hidden_state=hidden_states)
-
-
-class Qwen3_5ViTAsSiglipAdapter(nn.Module):
-    """SigLIP 契约 → NaViT 契约的翻译层。持有 Qwen3_5ViTBackbone。
-
-    对外暴露 SigLIP 式 forward(pixel_values=[N,3,H,W]) → BaseModelOutputWithPooling，
-    供 LlavaOnevision 消费；对内按官方 _preprocess 的 reshape 链把 pixel_values
-    转成 NaViT flat + grid_thw 喂给 Backbone。
-
-    reshape 链 1:1 照抄 transformers 官方 Qwen2VLImageProcessorFast._preprocess
-    (video_processing_qwen3_vl.py L227-252) —— Qwen3.5 无独立 image_processor，复用 Qwen3-VL 格式。
-    """
-
-    def __init__(self, vision_config: Qwen3_5VisionConfig):
-        super().__init__()
-        self.vision = Qwen3_5ViTBackbone(vision_config)
-        self.config = vision_config
-
-    @property
-    def dtype(self):
-        return next(self.parameters()).dtype
-
-    @property
-    def device(self):
-        return next(self.parameters()).device
-
-    def _flatten_navit(self, pixel_values: torch.Tensor):
-        """[N, 3, H, W] → (flat=[N*L, patch_dim], grid_thw=[N, 3], shape=(N, L)).
-
-        L = grid_t * grid_h * grid_w = 1 * (H/16) * (W/16)
-        patch_dim = C * temporal_patch_size * patch_size^2 = 3 * 2 * 16 * 16 = 1536
-        """
-        pixel_values = pixel_values.to(dtype=self.dtype)
-        tps = self.config.temporal_patch_size
-        ps = self.config.patch_size
-        ms = self.config.spatial_merge_size
-
-        patches = pixel_values.unsqueeze(1)
-        # 对单帧图像 T=1, pad=1 → expand 一帧使 T 整除 temporal_patch_size，
-        # Conv3d 在复制帧上退化为等效 2D Conv（数学无损）
-        T = patches.shape[1]
-        pad = -T % tps
-        if pad:
-            repeats = patches[:, -1:].expand(-1, pad, -1, -1, -1)
-            patches = torch.cat((patches, repeats), dim=1)
-
-        batch_size, t, channel, H, W = patches.shape
-        grid_t = t // tps
-        grid_h = H // ps
-        grid_w = W // ps
-
-        patches = patches.view(
-            batch_size, grid_t, tps, channel,
-            grid_h // ms, ms, ps,
-            grid_w // ms, ms, ps,
-        )
-        patches = patches.permute(0, 1, 4, 7, 5, 8, 3, 2, 6, 9)
-        flatten_patches = patches.reshape(
-            batch_size,
-            grid_t * grid_h * grid_w,
-            channel * tps * ps * ps,
-        )
-
-        seq_len = grid_t * grid_h * grid_w
-        flat = flatten_patches.reshape(batch_size * seq_len, -1)
-        # on-device 构造小 tensor 再 expand，host→GPU 同步量 O(3) 而非 O(N*3)
-        grid_unit = torch.tensor(
-            [grid_t, grid_h, grid_w], dtype=torch.int32, device=pixel_values.device,
-        )
-        grid_thw = grid_unit.unsqueeze(0).expand(batch_size, -1).contiguous()
-        return flat, grid_thw, (batch_size, seq_len)
-
-    def forward(
-        self,
-        pixel_values: torch.Tensor,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-        **kwargs,
-    ) -> BaseModelOutputWithPooling:
-        flat, grid_thw, (N, S) = self._flatten_navit(pixel_values)
-        vision_out = self.vision(flat, grid_thw=grid_thw)
-        hidden = vision_out.last_hidden_state.view(N, S, -1)
-
-        # Block-merge → row-major reorder before handing to LlavaOnevision.
-        # Internally the ViT runs in Qwen NaViT block-merge layout (pretrained
-        # pos_embed + RoPE contract); downstream LlavaOV `pack_image_features`
-        # (multi-tile AnyRes path, view(num_patch_h, num_patch_w, h, w, -1))
-        # and `apply_pooling` (video path, view(B, h, w, -1) + bilinear) BOTH
-        # assume row-major. Without this reorder, the multi-tile/video spatial
-        # pool pulls together tokens that are NOT spatially adjacent — silent
-        # corruption that doesn't fire on S1 single-tile path (line 348-351 of
-        # modeling_llava_onevision.py just flattens [N,D] verbatim) but kills
-        # S2 / eval quality.
-        grid_h = int(grid_thw[0, 1].item())
-        grid_w = int(grid_thw[0, 2].item())
-        ms = getattr(self.config, "spatial_merge_size", 2)
-        layout_perm = block_merge_to_row_major_permutation(
-            grid_h, grid_w, ms=ms, device=hidden.device,
-        )
-        hidden = hidden[:, layout_perm, :]
-
-        return BaseModelOutputWithPooling(
-            last_hidden_state=hidden,
-            # LlavaOnevision 索引 hidden_states[vision_feature_layer=-1]；tuple 长度 1 足够
-            hidden_states=(hidden,),
-            pooler_output=None,
-        )
-
-
-class LlavaQwen3_5ViTMultiModalProjector(nn.Module):
-    """标准 LlavaOnevision projector（V6 final_layernorm fix (2026-05-16): pre_norm → Identity）。
-
-    V6 final_layernorm fix (2026-05-16) 修复后, encoder 末端已自带 final_layernorm（与 SigLIP2 post_layernorm 对称），
-    projector 不再需要补 LN — pre_norm 改为 nn.Identity，对齐 SigLIP2 plugin 的
-    LlavaOnevision stock projector 结构（裸 linear_1 → GELU → linear_2），
-    保证 SigLIP2 / Qwen3.5 / Qwen3-VL 三个 backbone 在 LlavaOV 设定下公平对比。
-
-    （历史：V6.0.0~V6.0.4 时期 encoder 无 final LN，projector pre_norm 是补丁；
-    现在补丁回到 encoder 内部，projector 回归 stock 形态。）
-    """
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        super().__init__()
-        num_feature_layers = (
-            1 if isinstance(config.vision_feature_layer, int) else len(config.vision_feature_layer)
-        )
-        vision_dim = config.vision_config.hidden_size * num_feature_layers
-        text_dim = config.text_config.hidden_size
-        bias = getattr(config, "multimodal_projector_bias", True)
-
-        # V6 final_layernorm fix (2026-05-16): pre_norm = Identity (encoder 已自带 final_layernorm).
-        self.pre_norm = nn.Identity()
-        self.linear_1 = nn.Linear(vision_dim, text_dim, bias=bias)
-        self.act = ACT2FN[config.projector_hidden_act]
-        self.linear_2 = nn.Linear(text_dim, text_dim, bias=bias)
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        return self.linear_2(self.act(self.linear_1(self.pre_norm(x))))
-
-
-class LlavaQwen3_5ViTModel(LlavaOnevisionModel):
-    """继承 LlavaOnevisionModel 但绕过其 __init__ 手动装配。
-
-    父类 __init__ 调 `AutoModel.from_config(config.vision_config)` 会对 Qwen3_5VisionConfig
-    抛 "Unrecognized configuration"（Qwen3.5 vision 没注册到 AutoModel）。手动装配避开
-    这一步，同时省掉"先构造 Qwen3_5VisionModel 再被替换"的双重开销（~1.3GB init-peak）。
-
-    装配顺序与父类一致：vision_tower / projector / image_newline / language_model / post_init。
-    """
-
-    config_class = LlavaQwen3_5ViTConfig
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        # 跳过 LlavaOnevisionModel.__init__（AutoModel 不识别 Qwen3_5VisionConfig）
-        LlavaOnevisionPreTrainedModel.__init__(self, config)
-        self.vision_tower = Qwen3_5ViTAsSiglipAdapter(config.vision_config)
-        self.multi_modal_projector = LlavaQwen3_5ViTMultiModalProjector(config)
-        embed_std = 1 / math.sqrt(config.text_config.hidden_size)
-        self.image_newline = nn.Parameter(
-            torch.randn(config.text_config.hidden_size, dtype=self.dtype) * embed_std
-        )
-        self.vocab_size = config.text_config.vocab_size
-        self.language_model = AutoModel.from_config(config.text_config)
-        self.post_init()
-
-
-class LlavaQwen3_5ViTForConditionalGeneration(LlavaOnevisionForConditionalGeneration):
-    """继承 LlavaOnevisionForConditionalGeneration，只换 self.model。
-
-    同样跳过父类 __init__（避免重复构造 LlavaOnevisionModel，根因见 LlavaQwen3_5ViTModel）。
-    """
-
-    config_class = LlavaQwen3_5ViTConfig
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        LlavaOnevisionPreTrainedModel.__init__(self, config)
-        self.model = LlavaQwen3_5ViTModel(config)
-        self.lm_head = nn.Linear(
-            config.text_config.hidden_size, config.text_config.vocab_size, bias=False
-        )
-        self.post_init()
-
-
-AutoConfig.register(LlavaQwen3_5ViTConfig.model_type, LlavaQwen3_5ViTConfig)
-AutoModelForCausalLM.register(LlavaQwen3_5ViTConfig, LlavaQwen3_5ViTForConditionalGeneration)
-
-
-__all__ = [
-    "LlavaQwen3_5ViTConfig",
-    "Qwen3_5ViTBackbone",
-    "Qwen3_5ViTAsSiglipAdapter",
-    "LlavaQwen3_5ViTMultiModalProjector",
-    "LlavaQwen3_5ViTModel",
-    "LlavaQwen3_5ViTForConditionalGeneration",
-]
diff --git a/ckpts/4b_stock/processor_config.json b/ckpts/4b_stock/processor_config.json
deleted file mode 100644
index 204ed5d216c648cbd71da6c25d67b4bf1ec84433..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/processor_config.json
+++ /dev/null
@@ -1,203 +0,0 @@
-{
-  "image_processor": {
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_pad": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "image_grid_pinpoints": [
-      [
-        384,
-        384
-      ],
-      [
-        384,
-        768
-      ],
-      [
-        384,
-        1152
-      ],
-      [
-        384,
-        1536
-      ],
-      [
-        384,
-        1920
-      ],
-      [
-        384,
-        2304
-      ],
-      [
-        768,
-        384
-      ],
-      [
-        768,
-        768
-      ],
-      [
-        768,
-        1152
-      ],
-      [
-        768,
-        1536
-      ],
-      [
-        768,
-        1920
-      ],
-      [
-        768,
-        2304
-      ],
-      [
-        1152,
-        384
-      ],
-      [
-        1152,
-        768
-      ],
-      [
-        1152,
-        1152
-      ],
-      [
-        1152,
-        1536
-      ],
-      [
-        1152,
-        1920
-      ],
-      [
-        1152,
-        2304
-      ],
-      [
-        1536,
-        384
-      ],
-      [
-        1536,
-        768
-      ],
-      [
-        1536,
-        1152
-      ],
-      [
-        1536,
-        1536
-      ],
-      [
-        1536,
-        1920
-      ],
-      [
-        1536,
-        2304
-      ],
-      [
-        1920,
-        384
-      ],
-      [
-        1920,
-        768
-      ],
-      [
-        1920,
-        1152
-      ],
-      [
-        1920,
-        1536
-      ],
-      [
-        1920,
-        1920
-      ],
-      [
-        1920,
-        2304
-      ],
-      [
-        2304,
-        384
-      ],
-      [
-        2304,
-        768
-      ],
-      [
-        2304,
-        1152
-      ],
-      [
-        2304,
-        1536
-      ],
-      [
-        2304,
-        1920
-      ],
-      [
-        2304,
-        2304
-      ]
-    ],
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_processor_type": "LlavaOnevisionImageProcessor",
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "size": {
-      "height": 384,
-      "width": 384
-    }
-  },
-  "image_token": "<image>",
-  "num_image_tokens": 576,
-  "processor_class": "LlavaOnevisionProcessor",
-  "video_processor": {
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "do_sample_frames": false,
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "return_metadata": false,
-    "size": {
-      "height": 384,
-      "width": 384
-    },
-    "video_processor_type": "LlavaOnevisionVideoProcessor"
-  },
-  "video_token": "<video>",
-  "vision_aspect_ratio": "anyres_max_9",
-  "vision_feature_select_strategy": null
-}
diff --git a/ckpts/4b_stock/tokenizer.json b/ckpts/4b_stock/tokenizer.json
deleted file mode 100644
index c40d89c9226d696c215d65a68eea68c6dd55d79e..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/tokenizer.json
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:9024318c850eaadf26be79389d21b07a7afd8f1af749b89f9109b06c0d12173c
-size 11423018
diff --git a/ckpts/4b_stock/tokenizer_config.json b/ckpts/4b_stock/tokenizer_config.json
deleted file mode 100644
index 91a90671a687fb5d1fc3ed576d216e5e67d9e11b..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/tokenizer_config.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "add_prefix_space": false,
-  "backend": "tokenizers",
-  "bos_token": null,
-  "clean_up_tokenization_spaces": false,
-  "eos_token": "<|im_end|>",
-  "errors": "replace",
-  "extra_special_tokens": [
-    "<image>",
-    "<video>"
-  ],
-  "is_local": true,
-  "model_max_length": 1010000,
-  "pad_token": "<|endoftext|>",
-  "processor_class": "LlavaOnevisionProcessor",
-  "split_special_tokens": false,
-  "tokenizer_class": "Qwen2Tokenizer",
-  "unk_token": null
-}
diff --git a/ckpts/4b_stock/zero_to_fp32.py b/ckpts/4b_stock/zero_to_fp32.py
deleted file mode 100644
index 5995d6e6f04e43b989587aa9022a3aef0c66d694..0000000000000000000000000000000000000000
--- a/ckpts/4b_stock/zero_to_fp32.py
+++ /dev/null
@@ -1,760 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright (c) Microsoft Corporation.
-# SPDX-License-Identifier: Apache-2.0
-
-# DeepSpeed Team
-
-# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
-# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
-# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
-# application.
-#
-# example:
-#   python zero_to_fp32.py . output_dir/
-#   or
-#   python zero_to_fp32.py . output_dir/ --safe_serialization
-
-import argparse
-import torch
-import glob
-import math
-import os
-import re
-import gc
-import json
-import numpy as np
-from tqdm import tqdm
-from collections import OrderedDict
-from dataclasses import dataclass
-
-# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
-# DeepSpeed data structures it has to be available in the current python environment.
-from deepspeed.utils import logger
-from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
-                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
-                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
-
-
-@dataclass
-class zero_model_state:
-    buffers: dict()
-    param_shapes: dict()
-    shared_params: list
-    ds_version: int
-    frozen_param_shapes: dict()
-    frozen_param_fragments: dict()
-
-
-debug = 0
-
-# load to cpu
-device = torch.device('cpu')
-
-
-def atoi(text):
-    return int(text) if text.isdigit() else text
-
-
-def natural_keys(text):
-    '''
-    alist.sort(key=natural_keys) sorts in human order
-    http://nedbatchelder.com/blog/200712/human_sorting.html
-    (See Toothy's implementation in the comments)
-    '''
-    return [atoi(c) for c in re.split(r'(\d+)', text)]
-
-
-def get_model_state_file(checkpoint_dir, zero_stage):
-    if not os.path.isdir(checkpoint_dir):
-        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
-
-    # there should be only one file
-    if zero_stage <= 2:
-        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
-    elif zero_stage == 3:
-        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
-
-    if not os.path.exists(file):
-        raise FileNotFoundError(f"can't find model states file at '{file}'")
-
-    return file
-
-
-def get_checkpoint_files(checkpoint_dir, glob_pattern):
-    # XXX: need to test that this simple glob rule works for multi-node setup too
-    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
-
-    if len(ckpt_files) == 0:
-        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
-
-    return ckpt_files
-
-
-def get_optim_files(checkpoint_dir):
-    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
-
-
-def get_model_state_files(checkpoint_dir):
-    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
-
-
-def parse_model_states(files):
-    zero_model_states = []
-    for file in files:
-        state_dict = torch.load(file, map_location=device, weights_only=False)
-
-        if BUFFER_NAMES not in state_dict:
-            raise ValueError(f"{file} is not a model state checkpoint")
-        buffer_names = state_dict[BUFFER_NAMES]
-        if debug:
-            print("Found buffers:", buffer_names)
-
-        # recover just the buffers while restoring them to fp32 if they were saved in fp16
-        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
-        param_shapes = state_dict[PARAM_SHAPES]
-
-        # collect parameters that are included in param_shapes
-        param_names = []
-        for s in param_shapes:
-            for name in s.keys():
-                param_names.append(name)
-
-        # update with frozen parameters
-        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
-        if frozen_param_shapes is not None:
-            if debug:
-                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
-            param_names += list(frozen_param_shapes.keys())
-
-        # handle shared params
-        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
-
-        ds_version = state_dict.get(DS_VERSION, None)
-
-        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
-
-        z_model_state = zero_model_state(buffers=buffers,
-                                         param_shapes=param_shapes,
-                                         shared_params=shared_params,
-                                         ds_version=ds_version,
-                                         frozen_param_shapes=frozen_param_shapes,
-                                         frozen_param_fragments=frozen_param_fragments)
-        zero_model_states.append(z_model_state)
-
-    return zero_model_states
-
-
-def parse_optim_states(files, ds_checkpoint_dir):
-    total_files = len(files)
-    state_dicts = []
-    for f in tqdm(files, desc='Loading checkpoint shards'):
-        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
-        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
-        # and also handle the case where it was already removed by another helper script
-        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
-        state_dicts.append(state_dict)
-
-    if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
-        raise ValueError(f"{files[0]} is not a zero checkpoint")
-    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
-    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
-
-    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
-    # parameters can be different from data parallelism for non-expert parameters. So we can just
-    # use the max of the partition_count to get the dp world_size.
-
-    if type(world_size) is list:
-        world_size = max(world_size)
-
-    if world_size != total_files:
-        raise ValueError(
-            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
-            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
-        )
-
-    # the groups are named differently in each stage
-    if zero_stage <= 2:
-        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
-    elif zero_stage == 3:
-        fp32_groups_key = FP32_FLAT_GROUPS
-    else:
-        raise ValueError(f"unknown zero stage {zero_stage}")
-
-    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
-    return zero_stage, world_size, fp32_flat_groups
-
-
-def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
-    """
-    Returns fp32 state_dict reconstructed from ds checkpoint
-
-    Args:
-        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
-
-    """
-    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
-
-    optim_files = get_optim_files(ds_checkpoint_dir)
-    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
-    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
-
-    model_files = get_model_state_files(ds_checkpoint_dir)
-
-    zero_model_states = parse_model_states(model_files)
-    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
-
-    if zero_stage <= 2:
-        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                                          exclude_frozen_parameters)
-    elif zero_stage == 3:
-        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                                          exclude_frozen_parameters)
-
-
-def _zero2_merge_frozen_params(state_dict, zero_model_states):
-    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
-        return
-
-    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
-    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
-
-    if debug:
-        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
-        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
-
-        wanted_params = len(frozen_param_shapes)
-        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
-        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
-        print(f'Frozen params: Have {avail_numel} numels to process.')
-        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
-
-    total_params = 0
-    total_numel = 0
-    for name, shape in frozen_param_shapes.items():
-        total_params += 1
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-
-        state_dict[name] = frozen_param_fragments[name]
-
-        if debug:
-            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
-
-    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _has_callable(obj, fn):
-    attr = getattr(obj, fn, None)
-    return callable(attr)
-
-
-def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
-    param_shapes = zero_model_states[0].param_shapes
-
-    # Reconstruction protocol:
-    #
-    # XXX: document this
-
-    if debug:
-        for i in range(world_size):
-            for j in range(len(fp32_flat_groups[0])):
-                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
-
-    # XXX: memory usage doubles here (zero2)
-    num_param_groups = len(fp32_flat_groups[0])
-    merged_single_partition_of_fp32_groups = []
-    for i in range(num_param_groups):
-        merged_partitions = [sd[i] for sd in fp32_flat_groups]
-        full_single_fp32_vector = torch.cat(merged_partitions, 0)
-        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
-    avail_numel = sum(
-        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
-
-    if debug:
-        wanted_params = sum([len(shapes) for shapes in param_shapes])
-        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
-        # not asserting if there is a mismatch due to possible padding
-        print(f"Have {avail_numel} numels to process.")
-        print(f"Need {wanted_numel} numels in {wanted_params} params.")
-
-    # params
-    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
-    # out-of-core computing solution
-    total_numel = 0
-    total_params = 0
-    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
-        offset = 0
-        avail_numel = full_single_fp32_vector.numel()
-        for name, shape in shapes.items():
-
-            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
-            total_numel += unpartitioned_numel
-            total_params += 1
-
-            if debug:
-                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
-            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
-            offset += unpartitioned_numel
-
-        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
-        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
-        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
-        # live optimizer object, so we are checking that the numbers are within the right range
-        align_to = 2 * world_size
-
-        def zero2_align(x):
-            return align_to * math.ceil(x / align_to)
-
-        if debug:
-            print(f"original offset={offset}, avail_numel={avail_numel}")
-
-        offset = zero2_align(offset)
-        avail_numel = zero2_align(avail_numel)
-
-        if debug:
-            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
-
-        # Sanity check
-        if offset != avail_numel:
-            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
-
-    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                               exclude_frozen_parameters):
-    state_dict = OrderedDict()
-
-    # buffers
-    buffers = zero_model_states[0].buffers
-    state_dict.update(buffers)
-    if debug:
-        print(f"added {len(buffers)} buffers")
-
-    if not exclude_frozen_parameters:
-        _zero2_merge_frozen_params(state_dict, zero_model_states)
-
-    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
-
-    # recover shared parameters
-    for pair in zero_model_states[0].shared_params:
-        if pair[1] in state_dict:
-            state_dict[pair[0]] = state_dict[pair[1]]
-
-    return state_dict
-
-
-def zero3_partitioned_param_info(unpartitioned_numel, world_size):
-    remainder = unpartitioned_numel % world_size
-    padding_numel = (world_size - remainder) if remainder else 0
-    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
-    return partitioned_numel, padding_numel
-
-
-def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
-    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
-        return
-
-    if debug:
-        for i in range(world_size):
-            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
-            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
-
-        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
-        wanted_params = len(frozen_param_shapes)
-        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
-        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
-        print(f'Frozen params: Have {avail_numel} numels to process.')
-        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
-
-    total_params = 0
-    total_numel = 0
-    for name, shape in zero_model_states[0].frozen_param_shapes.items():
-        total_params += 1
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-
-        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
-        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
-
-        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
-
-        if debug:
-            print(
-                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
-            )
-
-    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
-
-
-class GatheredTensor:
-    """
-    A pseudo tensor that collects partitioned weights.
-    It is more memory efficient when there are multiple groups.
-    """
-
-    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
-        self.flat_groups = flat_groups
-        self.flat_groups_offset = flat_groups_offset
-        self.offset = offset
-        self.partitioned_numel = partitioned_numel
-        self.shape = shape
-        self.dtype = self.flat_groups[0][0].dtype
-
-    def contiguous(self):
-        """
-        Merge partitioned weights from flat_groups into a single tensor.
-        """
-        end_idx = self.offset + self.partitioned_numel
-        world_size = len(self.flat_groups)
-        pad_flat_param_chunks = []
-
-        for rank_i in range(world_size):
-            # for each rank, we need to collect weights from related group/groups
-            flat_groups_at_rank_i = self.flat_groups[rank_i]
-            start_group_id = None
-            end_group_id = None
-            for group_id in range(len(self.flat_groups_offset)):
-                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
-                    start_group_id = group_id
-                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
-                    end_group_id = group_id
-                    break
-            # collect weights from related group/groups
-            for group_id in range(start_group_id, end_group_id + 1):
-                flat_tensor = flat_groups_at_rank_i[group_id]
-                start_offset = self.offset - self.flat_groups_offset[group_id]
-                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
-                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
-
-        # collect weights from all ranks
-        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
-        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
-        return param
-
-
-def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
-    param_shapes = zero_model_states[0].param_shapes
-    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
-
-    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
-    # param, re-consolidating each param, while dealing with padding if any
-
-    # merge list of dicts, preserving order
-    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
-
-    if debug:
-        for i in range(world_size):
-            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
-
-        wanted_params = len(param_shapes)
-        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
-        # not asserting if there is a mismatch due to possible padding
-        avail_numel = fp32_flat_groups[0].numel() * world_size
-        print(f"Trainable params: Have {avail_numel} numels to process.")
-        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
-
-    # params
-    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
-    # out-of-core computing solution
-    offset = 0
-    total_numel = 0
-    total_params = 0
-    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
-    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-        total_params += 1
-        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
-
-        if debug:
-            print(
-                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
-            )
-
-        # memory efficient tensor
-        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
-        state_dict[name] = tensor
-        offset += partitioned_numel
-
-    offset *= world_size
-
-    # Sanity check
-    if offset != avail_numel:
-        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
-
-    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                               exclude_frozen_parameters):
-    state_dict = OrderedDict()
-
-    # buffers
-    buffers = zero_model_states[0].buffers
-    state_dict.update(buffers)
-    if debug:
-        print(f"added {len(buffers)} buffers")
-
-    if not exclude_frozen_parameters:
-        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
-
-    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
-
-    # recover shared parameters
-    for pair in zero_model_states[0].shared_params:
-        if pair[1] in state_dict:
-            state_dict[pair[0]] = state_dict[pair[1]]
-
-    return state_dict
-
-
-def to_torch_tensor(state_dict, return_empty_tensor=False):
-    """
-    Convert state_dict of GatheredTensor to torch tensor
-    """
-    torch_state_dict = {}
-    converted_tensors = {}
-    for name, tensor in state_dict.items():
-        tensor_id = id(tensor)
-        if tensor_id in converted_tensors:  # shared tensors
-            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
-            torch_state_dict[name] = shared_tensor
-        else:
-            converted_tensors[tensor_id] = name
-            if return_empty_tensor:
-                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
-            else:
-                torch_state_dict[name] = tensor.contiguous()
-    return torch_state_dict
-
-
-def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
-                                             tag=None,
-                                             exclude_frozen_parameters=False,
-                                             lazy_mode=False):
-    """
-    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
-    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
-    via a model hub.
-
-    Args:
-        - ``checkpoint_dir``: path to the desired checkpoint folder
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
-        - ``exclude_frozen_parameters``: exclude frozen parameters
-        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
-          Convert the pesduo tensor to torch tensor by ``.contiguous()``
-
-    Returns:
-        - pytorch ``state_dict``
-
-    A typical usage might be ::
-
-        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
-        # do the training and checkpoint saving
-        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
-        model = model.cpu() # move to cpu
-        model.load_state_dict(state_dict)
-        # submit to model hub or save the model to share with others
-
-    In this example the ``model`` will no longer be usable in the deepspeed context of the same
-    application. i.e. you will need to re-initialize the deepspeed engine, since
-    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
-
-    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
-
-    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
-    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
-    the checkpoint. Or you can load state_dict in lazy mode ::
-
-        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
-        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
-        for name, lazy_tensor in state_dict.item():
-            tensor = lazy_tensor.contiguous()  # to cpu
-            print(name, tensor)
-            # del tensor to release memory if it no longer in use
-    """
-    if tag is None:
-        latest_path = os.path.join(checkpoint_dir, 'latest')
-        if os.path.isfile(latest_path):
-            with open(latest_path, 'r') as fd:
-                tag = fd.read().strip()
-        else:
-            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
-
-    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
-
-    if not os.path.isdir(ds_checkpoint_dir):
-        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
-
-    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
-    if lazy_mode:
-        return state_dict
-    else:
-        return to_torch_tensor(state_dict)
-
-
-def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
-                                               output_dir,
-                                               max_shard_size="5GB",
-                                               safe_serialization=False,
-                                               tag=None,
-                                               exclude_frozen_parameters=False):
-    """
-    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
-    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
-
-    Args:
-        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
-        - ``output_dir``: directory to the pytorch fp32 state_dict output files
-        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
-        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
-        - ``exclude_frozen_parameters``: exclude frozen parameters
-    """
-
-    # Dependency pre-check
-    if safe_serialization:
-        try:
-            from safetensors.torch import save_file
-        except ImportError:
-            print('If you want to use `safe_serialization`, please `pip install safetensors`')
-            raise
-    if max_shard_size is not None:
-        try:
-            from huggingface_hub import split_torch_state_dict_into_shards
-        except ImportError:
-            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
-            raise
-
-    # Convert zero checkpoint to state_dict
-    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
-                                                          tag,
-                                                          exclude_frozen_parameters,
-                                                          lazy_mode=True)
-
-    # Shard the model if it is too big.
-    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
-    if max_shard_size is not None:
-        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
-        # an memory-efficient approach for sharding
-        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
-        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
-                                                              filename_pattern=filename_pattern,
-                                                              max_shard_size=max_shard_size)
-    else:
-        from collections import namedtuple
-        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
-        state_dict_split = StateDictSplit(is_sharded=False,
-                                          filename_to_tensors={weights_name: list(state_dict.keys())})
-
-    # Save the model by shard
-    os.makedirs(output_dir, exist_ok=True)
-    filename_to_tensors = state_dict_split.filename_to_tensors.items()
-    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
-        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
-        shard_state_dict = to_torch_tensor(shard_state_dict)
-        output_path = os.path.join(output_dir, shard_file)
-        if safe_serialization:
-            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
-        else:
-            torch.save(shard_state_dict, output_path)
-        # release the memory of current shard
-        for tensor_name in list(shard_state_dict.keys()):
-            del state_dict[tensor_name]
-            del shard_state_dict[tensor_name]
-        del shard_state_dict
-        gc.collect()
-
-    # Save index if sharded
-    if state_dict_split.is_sharded:
-        index = {
-            "metadata": state_dict_split.metadata,
-            "weight_map": state_dict_split.tensor_to_filename,
-        }
-        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
-        save_index_file = os.path.join(output_dir, save_index_file)
-        with open(save_index_file, "w", encoding="utf-8") as f:
-            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
-            f.write(content)
-
-
-def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
-    """
-    1. Put the provided model to cpu
-    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
-    3. Load it into the provided model
-
-    Args:
-        - ``model``: the model object to update
-        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
-
-    Returns:
-        - ``model`: modified model
-
-    Make sure you have plenty of CPU memory available before you call this function. If you don't
-    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
-    conveniently placed for you in the checkpoint folder.
-
-    A typical usage might be ::
-
-        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
-        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
-        # submit to model hub or save the model to share with others
-
-    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
-    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
-    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
-
-    """
-    logger.info("Extracting fp32 weights")
-    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
-
-    logger.info("Overwriting model with fp32 weights")
-    model = model.cpu()
-    model.load_state_dict(state_dict, strict=False)
-
-    return model
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument("checkpoint_dir",
-                        type=str,
-                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
-    parser.add_argument("output_dir",
-                        type=str,
-                        help="directory to the pytorch fp32 state_dict output files"
-                        "(e.g. path/checkpoint-12-output/)")
-    parser.add_argument(
-        "--max_shard_size",
-        type=str,
-        default="5GB",
-        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
-        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
-        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
-        "without CPU OOM issues.")
-    parser.add_argument(
-        "--safe_serialization",
-        default=False,
-        action='store_true',
-        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
-    parser.add_argument("-t",
-                        "--tag",
-                        type=str,
-                        default=None,
-                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
-    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
-    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
-    args = parser.parse_args()
-
-    debug = args.debug
-
-    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
-                                               args.output_dir,
-                                               max_shard_size=args.max_shard_size,
-                                               safe_serialization=args.safe_serialization,
-                                               tag=args.tag,
-                                               exclude_frozen_parameters=args.exclude_frozen_parameters)
diff --git a/ckpts/4b_v9_1/args.json b/ckpts/4b_v9_1/args.json
deleted file mode 100644
index 81ac86ccc525271d8c34b2d3417eda9d4c501d3e..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/args.json
+++ /dev/null
@@ -1,376 +0,0 @@
-{
-  "output_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_v9_1_a800_10pct/v0-20260602-181529",
-  "per_device_train_batch_size": 1,
-  "num_train_epochs": 3.0,
-  "max_steps": 505,
-  "learning_rate": 1e-05,
-  "lr_scheduler_type": "cosine",
-  "lr_scheduler_kwargs": null,
-  "warmup_steps": 0,
-  "optim": "adamw_torch_fused",
-  "optim_args": null,
-  "weight_decay": 0.1,
-  "adam_beta1": 0.9,
-  "adam_beta2": 0.95,
-  "adam_epsilon": 1e-08,
-  "optim_target_modules": null,
-  "gradient_accumulation_steps": 8,
-  "average_tokens_across_devices": true,
-  "max_grad_norm": 1.0,
-  "label_smoothing_factor": 0.0,
-  "bf16": true,
-  "fp16": false,
-  "bf16_full_eval": false,
-  "fp16_full_eval": false,
-  "tf32": null,
-  "gradient_checkpointing": true,
-  "gradient_checkpointing_kwargs": "{\"use_reentrant\": false}",
-  "torch_compile": false,
-  "torch_compile_backend": null,
-  "torch_compile_mode": null,
-  "use_liger_kernel": false,
-  "liger_kernel_config": null,
-  "use_cache": false,
-  "neftune_noise_alpha": null,
-  "torch_empty_cache_steps": null,
-  "auto_find_batch_size": false,
-  "logging_strategy": "steps",
-  "logging_steps": 1,
-  "logging_first_step": true,
-  "log_on_each_node": true,
-  "logging_nan_inf_filter": true,
-  "include_num_input_tokens_seen": false,
-  "log_level": "passive",
-  "log_level_replica": "warning",
-  "disable_tqdm": null,
-  "report_to": [
-    "none"
-  ],
-  "run_name": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_v9_1_a800_10pct/v0-20260602-181529",
-  "project": "huggingface",
-  "trackio_space_id": "trackio",
-  "eval_strategy": "no",
-  "eval_steps": 200.0,
-  "eval_delay": 0,
-  "per_device_eval_batch_size": 1,
-  "prediction_loss_only": false,
-  "eval_on_start": false,
-  "eval_do_concat_batches": true,
-  "eval_use_gather_object": false,
-  "eval_accumulation_steps": null,
-  "include_for_metrics": [],
-  "batch_eval_metrics": false,
-  "save_only_model": false,
-  "save_strategy": "steps",
-  "save_steps": 200.0,
-  "save_on_each_node": false,
-  "save_total_limit": 1,
-  "enable_jit_checkpoint": false,
-  "push_to_hub": false,
-  "hub_token": null,
-  "hub_private_repo": null,
-  "hub_model_id": null,
-  "hub_strategy": "every_save",
-  "hub_always_push": false,
-  "hub_revision": null,
-  "load_best_model_at_end": false,
-  "metric_for_best_model": "loss",
-  "greater_is_better": false,
-  "ignore_data_skip": false,
-  "restore_callback_states_from_checkpoint": false,
-  "full_determinism": false,
-  "seed": 42,
-  "data_seed": 42,
-  "use_cpu": false,
-  "accelerator_config": "{\"dispatch_batches\": false}",
-  "parallelism_config": null,
-  "dataloader_drop_last": false,
-  "dataloader_num_workers": 6,
-  "dataloader_pin_memory": true,
-  "dataloader_persistent_workers": true,
-  "dataloader_prefetch_factor": 4,
-  "remove_unused_columns": true,
-  "label_names": null,
-  "train_sampling_strategy": "random",
-  "length_column_name": "length",
-  "ddp_find_unused_parameters": null,
-  "ddp_bucket_cap_mb": null,
-  "ddp_broadcast_buffers": null,
-  "ddp_backend": null,
-  "ddp_timeout": 7200,
-  "fsdp": [],
-  "fsdp_config": null,
-  "deepspeed": {
-    "fp16": {
-      "enabled": "auto",
-      "loss_scale": 0,
-      "loss_scale_window": 1000,
-      "initial_scale_power": 16,
-      "hysteresis": 2,
-      "min_loss_scale": 1
-    },
-    "bf16": {
-      "enabled": "auto"
-    },
-    "zero_optimization": {
-      "stage": 2,
-      "offload_optimizer": {
-        "device": "none",
-        "pin_memory": true
-      },
-      "allgather_partitions": true,
-      "allgather_bucket_size": 200000000.0,
-      "overlap_comm": false,
-      "reduce_scatter": true,
-      "reduce_bucket_size": 200000000.0,
-      "contiguous_gradients": true
-    },
-    "gradient_accumulation_steps": "auto",
-    "gradient_clipping": "auto",
-    "steps_per_print": 2000,
-    "train_batch_size": "auto",
-    "train_micro_batch_size_per_gpu": "auto",
-    "wall_clock_breakdown": false
-  },
-  "debug": null,
-  "skip_memory_metrics": true,
-  "do_train": false,
-  "do_eval": false,
-  "do_predict": false,
-  "resume_from_checkpoint": null,
-  "warmup_ratio": 0.05,
-  "logging_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_v9_1_a800_10pct/v0-20260602-181529/runs",
-  "local_rank": 0,
-  "sortish_sampler": false,
-  "predict_with_generate": false,
-  "generation_max_length": null,
-  "generation_num_beams": null,
-  "generation_config": null,
-  "tuner_backend": "peft",
-  "vit_gradient_checkpointing": true,
-  "router_aux_loss_coef": 0.0,
-  "enable_dft_loss": false,
-  "enable_channel_loss": false,
-  "safe_serialization": true,
-  "max_shard_size": "5GB",
-  "check_model": true,
-  "acc_strategy": "token",
-  "train_dataloader_shuffle": true,
-  "group_by_length": false,
-  "max_epochs": null,
-  "aligner_lr": null,
-  "vit_lr": 1e-06,
-  "use_logits_to_keep": null,
-  "ds3_gather_for_generation": true,
-  "resume_only_model": false,
-  "optimizer": null,
-  "loss_type": null,
-  "eval_metric": null,
-  "callbacks": [],
-  "early_stop_interval": null,
-  "eval_use_evalscope": false,
-  "eval_dataset": [],
-  "eval_dataset_args": null,
-  "eval_limit": null,
-  "eval_generation_config": null,
-  "extra_eval_args": null,
-  "tuner_type": "full",
-  "use_galore": false,
-  "galore_target_modules": null,
-  "galore_rank": 128,
-  "galore_update_proj_gap": 50,
-  "galore_scale": 1.0,
-  "galore_proj_type": "std",
-  "galore_optim_per_parameter": false,
-  "galore_with_embedding": false,
-  "galore_quantization": false,
-  "galore_proj_quant": false,
-  "galore_proj_bits": 4,
-  "galore_proj_group_size": 256,
-  "galore_cos_threshold": 0.4,
-  "galore_gamma_proj": 2,
-  "galore_queue_size": 5,
-  "lisa_activated_layers": 0,
-  "lisa_step_interval": 20,
-  "use_flash_ckpt": false,
-  "use_ray": false,
-  "ray_exp_name": null,
-  "device_groups": null,
-  "model": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_v9_1_a800/v0-20260602-115055/checkpoint-2181",
-  "model_type": "llava_qwen3_5vit_qwen3",
-  "model_revision": null,
-  "task_type": "causal_lm",
-  "torch_dtype": "bfloat16",
-  "attn_impl": "flash_attn",
-  "experts_impl": null,
-  "new_special_tokens": [],
-  "num_labels": null,
-  "problem_type": null,
-  "rope_scaling": null,
-  "device_map": null,
-  "max_memory": {},
-  "max_model_len": null,
-  "local_repo_path": null,
-  "init_strategy": null,
-  "template": "llava_qwen3_5vit_qwen3",
-  "system": null,
-  "max_length": 16384,
-  "truncation_strategy": "delete",
-  "max_pixels": null,
-  "agent_template": null,
-  "norm_bbox": null,
-  "use_chat_template": true,
-  "padding_side": "right",
-  "padding_free": true,
-  "loss_scale": "default",
-  "sequence_parallel_size": 1,
-  "template_backend": "swift",
-  "response_prefix": null,
-  "enable_thinking": null,
-  "add_non_thinking_prefix": true,
-  "dataset": [],
-  "val_dataset": [],
-  "cached_dataset": [
-    "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/image_10pct/train",
-    "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/video_10pct/train"
-  ],
-  "cached_val_dataset": [],
-  "split_dataset_ratio": 0.0,
-  "dataset_num_proc": 16,
-  "load_from_cache_file": false,
-  "dataset_shuffle": true,
-  "val_dataset_shuffle": false,
-  "streaming": false,
-  "interleave_prob": null,
-  "stopping_strategy": "first_exhausted",
-  "shuffle_buffer_size": 1000,
-  "download_mode": "reuse_dataset_if_exists",
-  "columns": {},
-  "strict": false,
-  "model_name": null,
-  "model_author": null,
-  "custom_dataset_info": [],
-  "quant_method": null,
-  "quant_bits": null,
-  "hqq_axis": null,
-  "bnb_4bit_compute_dtype": "bfloat16",
-  "bnb_4bit_quant_type": "nf4",
-  "bnb_4bit_use_double_quant": true,
-  "bnb_4bit_quant_storage": null,
-  "max_new_tokens": 64,
-  "temperature": 0.0,
-  "top_k": null,
-  "top_p": null,
-  "repetition_penalty": null,
-  "num_beams": 1,
-  "stream": false,
-  "stop_words": [],
-  "logprobs": false,
-  "top_logprobs": null,
-  "structured_outputs_regex": null,
-  "adapters": [],
-  "external_plugins": [
-    "ms-swift/video_mllm/model_plugin_qwen3_5vit.py",
-    "ms-swift/video_mllm/dataset_plugin.py"
-  ],
-  "custom_register_path": [],
-  "model_kwargs": {},
-  "load_args": false,
-  "load_data_args": false,
-  "packing": true,
-  "packing_length": 16384,
-  "packing_num_proc": 1,
-  "lazy_tokenize": false,
-  "use_hf": true,
-  "ignore_args_error": false,
-  "use_swift_lora": false,
-  "freeze_parameters": [],
-  "freeze_parameters_regex": null,
-  "freeze_parameters_ratio": 0.0,
-  "trainable_parameters": [
-    "model.multi_modal_projector"
-  ],
-  "trainable_parameters_regex": null,
-  "freeze_llm": false,
-  "freeze_vit": false,
-  "freeze_aligner": false,
-  "target_modules": [
-    "all-linear"
-  ],
-  "target_regex": null,
-  "target_parameters": null,
-  "modules_to_save": [],
-  "lora_rank": 8,
-  "lora_alpha": 32,
-  "lora_dropout": 0.05,
-  "lora_bias": "none",
-  "lora_dtype": null,
-  "lorap_lr_ratio": null,
-  "use_rslora": false,
-  "use_dora": false,
-  "lora_ga_batch_size": 2,
-  "lora_ga_iters": 2,
-  "lora_ga_max_length": 1024,
-  "lora_ga_direction": "ArB2r",
-  "lora_ga_scale": "stable",
-  "lora_ga_stable_gamma": 16,
-  "init_weights": true,
-  "fourier_n_frequency": 2000,
-  "fourier_scaling": 300.0,
-  "boft_block_size": 4,
-  "boft_block_num": 0,
-  "boft_n_butterfly_factor": 1,
-  "boft_dropout": 0.0,
-  "vera_rank": 256,
-  "vera_projection_prng_key": 0,
-  "vera_dropout": 0.0,
-  "vera_d_initial": 0.1,
-  "adapter_act": "gelu",
-  "adapter_length": 128,
-  "adalora_target_r": 8,
-  "adalora_init_r": 12,
-  "adalora_tinit": 0,
-  "adalora_tfinal": 0,
-  "adalora_deltaT": 1,
-  "adalora_beta1": 0.85,
-  "adalora_beta2": 0.85,
-  "adalora_orth_reg_weight": 0.5,
-  "llamapro_num_new_blocks": 4,
-  "llamapro_num_groups": null,
-  "reft_layer_key": null,
-  "reft_layers": null,
-  "reft_rank": 4,
-  "reft_intervention_type": "LoreftIntervention",
-  "reft_args": null,
-  "swanlab_token": null,
-  "swanlab_project": "ms-swift",
-  "swanlab_workspace": null,
-  "swanlab_exp_name": null,
-  "swanlab_notification_method": null,
-  "swanlab_webhook_url": null,
-  "swanlab_secret": null,
-  "swanlab_sender_email": null,
-  "swanlab_receiver_email": null,
-  "swanlab_smtp_server": null,
-  "swanlab_smtp_port": null,
-  "swanlab_email_language": "zh",
-  "swanlab_mode": "cloud",
-  "add_version": true,
-  "create_checkpoint_symlink": false,
-  "zero_hpz_partition_size": null,
-  "deepspeed_autotp_size": null,
-  "swift_version": "4.1.1",
-  "ckpt_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_v9_1_a800/v0-20260602-115055/checkpoint-2181",
-  "rank": 0,
-  "global_world_size": 8,
-  "local_world_size": 8,
-  "model_suffix": "checkpoint-2181",
-  "model_info": "ModelInfo(model_type='llava_qwen3_5vit_qwen3', model_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_v9_1_a800/v0-20260602-115055/checkpoint-2181', torch_dtype=torch.bfloat16, max_model_len=262144, quant_method=None, quant_bits=None, rope_scaling=None, is_moe_model=False, is_multimodal=True, config=None, task_type='causal_lm', num_labels=None)",
-  "model_meta": "ModelMeta(model_type='llava_qwen3_5vit_qwen3', model_groups=[ModelGroup(models=[Model(ms_model_id='Qwen/Qwen3-0.6B', hf_model_id='Qwen/Qwen3-0.6B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-1.7B', hf_model_id='Qwen/Qwen3-1.7B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-4B', hf_model_id='Qwen/Qwen3-4B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-8B', hf_model_id='Qwen/Qwen3-8B', model_path=None, ms_revision=None, hf_revision=None)], template=None, ignore_patterns=None, requires=None, tags=[])], loader=<class 'model_plugin_qwen3_5vit.Qwen3_5ViTQwen3Loader'>, template='llava_qwen3_5vit_qwen3', model_arch=MultiModelKeys(arch_name='llava_hf', embedding=None, module_list=None, lm_head=None, q_proj=None, k_proj=None, v_proj=None, o_proj=None, attention=None, mlp=None, down_proj=None, qkv_proj=None, qk_proj=None, qa_proj=None, qb_proj=None, kv_proj=None, kva_proj=None, kvb_proj=None, language_model=['model.language_model', 'lm_head'], aligner=['model.multi_modal_projector'], vision_tower=['model.vision_tower'], generator=[]), mcore_model_type=None, architectures=['LlavaQwen3_5ViTForConditionalGeneration'], additional_saved_files=[], torch_dtype=None, is_multimodal=True, is_reward=False, task_type=None, ignore_patterns=None, requires=[], tags=['vision', 'video'])",
-  "model_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_v9_1_a800/v0-20260602-115055/checkpoint-2181",
-  "template_meta": "QwenTemplateMeta(template_type='llava_qwen3_5vit_qwen3', prefix=[], prompt=['<|im_start|>user\\n{{QUERY}}<|im_end|>\\n<|im_start|>assistant\\n'], chat_sep=['<|im_end|>\\n'], suffix=['<|im_end|>\\n'], template_cls=<class 'model_plugin_qwen3_5vit.Qwen3_5ViTLlavaTemplate'>, system_prefix=['<|im_start|>system\\n{{SYSTEM}}<|im_end|>\\n'], default_system=None, auto_add_bos=False, stop_words=['<|endoftext|>'], agent_template='hermes', is_thinking=False, thinking_prefix='', non_thinking_prefix='', history_thinking_prefix='')",
-  "_val_dataset_exists": false,
-  "hub": "<class 'swift.hub.hub.HFHub'>",
-  "evaluation_strategy": "steps",
-  "training_args": "Seq2SeqTrainingArguments(output_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_v9_1_a800_10pct/v0-20260602-181529', per_device_train_batch_size=1, num_train_epochs=3.0, max_steps=505, learning_rate=1e-05, lr_scheduler_type=<SchedulerType.COSINE: 'cosine'>, lr_scheduler_kwargs=None, warmup_steps=0.05, optim=<OptimizerNames.ADAMW_TORCH_FUSED: 'adamw_torch_fused'>, optim_args=None, weight_decay=0.1, adam_beta1=0.9, adam_beta2=0.95, adam_epsilon=1e-08, optim_target_modules=None, gradient_accumulation_steps=8, average_tokens_across_devices=None, max_grad_norm=1.0, label_smoothing_factor=0.0, bf16=True, fp16=False, bf16_full_eval=False, fp16_full_eval=False, tf32=None, gradient_checkpointing=True, gradient_checkpointing_kwargs={'use_reentrant': False}, torch_compile=False, torch_compile_backend=None, torch_compile_mode=None, use_liger_kernel=False, liger_kernel_config=None, use_cache=False, neftune_noise_alpha=None, torch_empty_cache_steps=None, auto_find_batch_size=False, logging_strategy=<IntervalStrategy.STEPS: 'steps'>, logging_steps=1, logging_first_step=True, log_on_each_node=True, logging_nan_inf_filter=True, include_num_input_tokens_seen=None, log_level='passive', log_level_replica='warning', disable_tqdm=False, report_to=[], run_name='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_v9_1_a800_10pct/v0-20260602-181529', project='huggingface', trackio_space_id='trackio', eval_strategy=<IntervalStrategy.NO: 'no'>, eval_steps=200.0, eval_delay=0, per_device_eval_batch_size=1, prediction_loss_only=False, eval_on_start=False, eval_do_concat_batches=True, eval_use_gather_object=False, eval_accumulation_steps=None, include_for_metrics=[], batch_eval_metrics=False, save_only_model=False, save_strategy=<SaveStrategy.STEPS: 'steps'>, save_steps=200, save_on_each_node=False, save_total_limit=1, enable_jit_checkpoint=False, push_to_hub=False, hub_token=None, hub_private_repo=None, hub_model_id=None, hub_strategy=<HubStrategy.EVERY_SAVE: 'every_save'>, hub_always_push=False, hub_revision=None, load_best_model_at_end=False, metric_for_best_model='loss', greater_is_better=False, ignore_data_skip=False, restore_callback_states_from_checkpoint=False, full_determinism=False, seed=42, data_seed=42, use_cpu=False, accelerator_config=AcceleratorConfig(split_batches=False, dispatch_batches=False, even_batches=True, use_seedable_sampler=True, non_blocking=False, gradient_accumulation_kwargs=None, use_configured_state=False), parallelism_config=None, dataloader_drop_last=False, dataloader_num_workers=6, dataloader_pin_memory=True, dataloader_persistent_workers=True, dataloader_prefetch_factor=4, remove_unused_columns=False, label_names=None, train_sampling_strategy='random', length_column_name='length', ddp_find_unused_parameters=None, ddp_bucket_cap_mb=None, ddp_broadcast_buffers=None, ddp_backend=None, ddp_timeout=7200, fsdp=[], fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}, deepspeed={'fp16': {'enabled': 'auto', 'loss_scale': 0, 'loss_scale_window': 1000, 'initial_scale_power': 16, 'hysteresis': 2, 'min_loss_scale': 1}, 'bf16': {'enabled': 'auto'}, 'zero_optimization': {'stage': 2, 'offload_optimizer': {'device': 'none', 'pin_memory': True}, 'allgather_partitions': True, 'allgather_bucket_size': 200000000.0, 'overlap_comm': False, 'reduce_scatter': True, 'reduce_bucket_size': 200000000.0, 'contiguous_gradients': True}, 'gradient_accumulation_steps': 'auto', 'gradient_clipping': 'auto', 'steps_per_print': 2000, 'train_batch_size': 'auto', 'train_micro_batch_size_per_gpu': 'auto', 'wall_clock_breakdown': False}, debug=[], skip_memory_metrics=True, do_train=False, do_eval=False, do_predict=False, resume_from_checkpoint=None, warmup_ratio=0.05, logging_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_v9_1_a800_10pct/v0-20260602-181529/runs', local_rank=0, sortish_sampler=False, predict_with_generate=False, generation_max_length=None, generation_num_beams=None, generation_config=None, tuner_backend='peft', vit_gradient_checkpointing=True, router_aux_loss_coef=0.0, enable_dft_loss=False, enable_channel_loss=False, safe_serialization=True, max_shard_size='5GB', check_model=True, acc_strategy='token', train_dataloader_shuffle=True, group_by_length=False, max_epochs=None, aligner_lr=None, vit_lr=1e-06, use_logits_to_keep=None, ds3_gather_for_generation=True, resume_only_model=False, optimizer='multimodal', loss_type=None, eval_metric=None, callbacks=[], early_stop_interval=None, eval_use_evalscope=False, eval_dataset=[], eval_dataset_args=None, eval_limit=None, eval_generation_config=None, extra_eval_args=None, tuner_type='full', use_galore=False, galore_target_modules=None, galore_rank=128, galore_update_proj_gap=50, galore_scale=1.0, galore_proj_type='std', galore_optim_per_parameter=False, galore_with_embedding=False, galore_quantization=False, galore_proj_quant=False, galore_proj_bits=4, galore_proj_group_size=256, galore_cos_threshold=0.4, galore_gamma_proj=2, galore_queue_size=5, lisa_activated_layers=0, lisa_step_interval=20, use_flash_ckpt=False)"
-}
\ No newline at end of file
diff --git a/ckpts/4b_v9_1/chat_template.jinja b/ckpts/4b_v9_1/chat_template.jinja
deleted file mode 100644
index 70adff8a08fb31e0636f618564838d4bf3c05286..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/chat_template.jinja
+++ /dev/null
@@ -1,61 +0,0 @@
-{%- if tools %}
-    {{- '<|im_start|>system\n' }}
-    {%- if messages[0].role == 'system' %}
-        {{- messages[0].content + '\n\n' }}
-    {%- endif %}
-    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
-    {%- for tool in tools %}
-        {{- "\n" }}
-        {{- tool | tojson }}
-    {%- endfor %}
-    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
-{%- else %}
-    {%- if messages[0].role == 'system' %}
-        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
-    {%- endif %}
-{%- endif %}
-{%- for message in messages %}
-    {%- if message.content is string %}
-        {%- set content = message.content %}
-    {%- else %}
-        {%- set content = '' %}
-    {%- endif %}
-    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
-        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
-    {%- elif message.role == "assistant" %}
-        {{- '<|im_start|>' + message.role + '\n' + content }}
-        {%- if message.tool_calls %}
-            {%- for tool_call in message.tool_calls %}
-                {%- if (loop.first and content) or (not loop.first) %}
-                    {{- '\n' }}
-                {%- endif %}
-                {%- if tool_call.function %}
-                    {%- set tool_call = tool_call.function %}
-                {%- endif %}
-                {{- '<tool_call>\n{"name": "' }}
-                {{- tool_call.name }}
-                {{- '", "arguments": ' }}
-                {%- if tool_call.arguments is string %}
-                    {{- tool_call.arguments }}
-                {%- else %}
-                    {{- tool_call.arguments | tojson }}
-                {%- endif %}
-                {{- '}\n</tool_call>' }}
-            {%- endfor %}
-        {%- endif %}
-        {{- '<|im_end|>\n' }}
-    {%- elif message.role == "tool" %}
-        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
-            {{- '<|im_start|>user' }}
-        {%- endif %}
-        {{- '\n<tool_response>\n' }}
-        {{- content }}
-        {{- '\n</tool_response>' }}
-        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
-            {{- '<|im_end|>\n' }}
-        {%- endif %}
-    {%- endif %}
-{%- endfor %}
-{%- if add_generation_prompt %}
-    {{- '<|im_start|>assistant\n' }}
-{%- endif %}
\ No newline at end of file
diff --git a/ckpts/4b_v9_1/chat_template.json b/ckpts/4b_v9_1/chat_template.json
deleted file mode 100644
index 68724c68b1612e9ab2e31f84e3279099ad656ff6..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/chat_template.json
+++ /dev/null
@@ -1,3 +0,0 @@
-{
-  "chat_template": "{% for message in messages %}{{'<|im_start|>' + message['role'] + ' '}}{# Render all images first #}{% for content in message['content'] | selectattr('type', 'equalto', 'image') %}{{ '<image>' }}{% endfor %}{# Render all video then #}{% for content in message['content'] | selectattr('type', 'equalto', 'video') %}{{ '<video>' }}{% endfor %}{# Render all text next #}{% if message['role'] != 'assistant' %}{% for content in message['content'] | selectattr('type', 'equalto', 'text') %}{{ '\n' + content['text'] }}{% endfor %}{% else %}{% for content in message['content'] | selectattr('type', 'equalto', 'text') %}{% generation %}{{ '\n' + content['text'] }}{% endgeneration %}{% endfor %}{% endif %}{{'<|im_end|>'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"
-}
diff --git a/ckpts/4b_v9_1/config.json b/ckpts/4b_v9_1/config.json
deleted file mode 100644
index ce830f6986f152f6aeb43595f9a789a81c62b646..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/config.json
+++ /dev/null
@@ -1,262 +0,0 @@
-{
-  "architectures": [
-    "LlavaQwen3_5ViTForConditionalGeneration"
-  ],
-  "bos_token_id": null,
-  "dtype": "bfloat16",
-  "eos_token_id": 151645,
-  "hidden_size": 2560,
-  "image_grid_pinpoints": [
-    [
-      384,
-      384
-    ],
-    [
-      384,
-      768
-    ],
-    [
-      384,
-      1152
-    ],
-    [
-      384,
-      1536
-    ],
-    [
-      384,
-      1920
-    ],
-    [
-      384,
-      2304
-    ],
-    [
-      768,
-      384
-    ],
-    [
-      768,
-      768
-    ],
-    [
-      768,
-      1152
-    ],
-    [
-      768,
-      1536
-    ],
-    [
-      768,
-      1920
-    ],
-    [
-      768,
-      2304
-    ],
-    [
-      1152,
-      384
-    ],
-    [
-      1152,
-      768
-    ],
-    [
-      1152,
-      1152
-    ],
-    [
-      1152,
-      1536
-    ],
-    [
-      1152,
-      1920
-    ],
-    [
-      1152,
-      2304
-    ],
-    [
-      1536,
-      384
-    ],
-    [
-      1536,
-      768
-    ],
-    [
-      1536,
-      1152
-    ],
-    [
-      1536,
-      1536
-    ],
-    [
-      1536,
-      1920
-    ],
-    [
-      1536,
-      2304
-    ],
-    [
-      1920,
-      384
-    ],
-    [
-      1920,
-      768
-    ],
-    [
-      1920,
-      1152
-    ],
-    [
-      1920,
-      1536
-    ],
-    [
-      1920,
-      1920
-    ],
-    [
-      1920,
-      2304
-    ],
-    [
-      2304,
-      384
-    ],
-    [
-      2304,
-      768
-    ],
-    [
-      2304,
-      1152
-    ],
-    [
-      2304,
-      1536
-    ],
-    [
-      2304,
-      1920
-    ],
-    [
-      2304,
-      2304
-    ]
-  ],
-  "image_token_index": 151669,
-  "model_type": "llava_qwen3_5vit_qwen3",
-  "multimodal_projector_bias": true,
-  "pad_token_id": 151643,
-  "projector_hidden_act": "gelu",
-  "text_config": {
-    "_name_or_path": "/share/m2v_intern_v3/wangjunjie09/model_cache/huggingface/Qwen/Qwen3-4B-Instruct-2507",
-    "architectures": [
-      "Qwen3ForCausalLM"
-    ],
-    "attention_bias": false,
-    "attention_dropout": 0.0,
-    "bos_token_id": 151643,
-    "dtype": "bfloat16",
-    "eos_token_id": 151645,
-    "head_dim": 128,
-    "hidden_act": "silu",
-    "hidden_size": 2560,
-    "initializer_range": 0.02,
-    "intermediate_size": 9728,
-    "layer_types": [
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention"
-    ],
-    "max_position_embeddings": 262144,
-    "max_window_layers": 36,
-    "model_type": "qwen3",
-    "num_attention_heads": 32,
-    "num_hidden_layers": 36,
-    "num_key_value_heads": 8,
-    "pad_token_id": 151643,
-    "rms_norm_eps": 1e-06,
-    "rope_parameters": {
-      "rope_theta": 5000000,
-      "rope_type": "default"
-    },
-    "sliding_window": null,
-    "tie_word_embeddings": true,
-    "use_cache": false,
-    "use_sliding_window": false,
-    "vocab_size": 151936
-  },
-  "tie_word_embeddings": true,
-  "transformers_version": "5.5.4",
-  "use_cache": false,
-  "video_token_index": 151670,
-  "vision_aspect_ratio": "anyres_max_9",
-  "vision_config": {
-    "deepstack_visual_indexes": [],
-    "depth": 24,
-    "dtype": "bfloat16",
-    "hidden_act": "gelu_pytorch_tanh",
-    "hidden_size": 1024,
-    "image_size": 384,
-    "in_channels": 3,
-    "initializer_range": 0.02,
-    "intermediate_size": 4096,
-    "model_type": "qwen3_5",
-    "num_heads": 16,
-    "num_position_embeddings": 2304,
-    "out_hidden_size": 2048,
-    "patch_size": 16,
-    "spatial_merge_size": 2,
-    "temporal_patch_size": 2
-  },
-  "vision_feature_layer": -1,
-  "vision_feature_select_strategy": "full",
-  "auto_map": {
-    "AutoConfig": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTConfig",
-    "AutoModel": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
-    "AutoModelForCausalLM": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
-    "AutoModelForImageTextToText": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration"
-  }
-}
\ No newline at end of file
diff --git a/ckpts/4b_v9_1/generation_config.json b/ckpts/4b_v9_1/generation_config.json
deleted file mode 100644
index 56320d3179df3512d0cb76c240831a10e6f7c6f7..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/generation_config.json
+++ /dev/null
@@ -1,12 +0,0 @@
-{
-  "_from_model_config": true,
-  "bos_token_id": 151643,
-  "eos_token_id": [
-    151645,
-    151643
-  ],
-  "output_attentions": false,
-  "output_hidden_states": false,
-  "transformers_version": "5.5.4",
-  "use_cache": true
-}
diff --git a/ckpts/4b_v9_1/latest b/ckpts/4b_v9_1/latest
deleted file mode 100644
index bda446cd4c4d586af0956cf22fb3bd54d8006aec..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/latest
+++ /dev/null
@@ -1 +0,0 @@
-global_step505
\ No newline at end of file
diff --git a/ckpts/4b_v9_1/model-00001-of-00002.safetensors b/ckpts/4b_v9_1/model-00001-of-00002.safetensors
deleted file mode 100644
index 16929854d6331c125bacebc4e0730fd2750b9f10..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/model-00001-of-00002.safetensors
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:daecd183bea10424f8e65aad0a672571b19b72fd25499991ff5e494c2072a448
-size 4987507408
diff --git a/ckpts/4b_v9_1/model-00002-of-00002.safetensors b/ckpts/4b_v9_1/model-00002-of-00002.safetensors
deleted file mode 100644
index 48019b586bbbd82d323f80a8200bcf1b75ab778c..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/model-00002-of-00002.safetensors
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:c36c9a449fbcb3cecc3ae61966ed4c49f1346a1543193ba3b00bf848b20153bd
-size 4466285248
diff --git a/ckpts/4b_v9_1/model.safetensors.index.json b/ckpts/4b_v9_1/model.safetensors.index.json
deleted file mode 100644
index a227833b5ba2c2ae9deeddde8d4fc716babffc8c..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/model.safetensors.index.json
+++ /dev/null
@@ -1,705 +0,0 @@
-{
-  "metadata": {
-    "total_parameters": 4337895424,
-    "total_size": 9453703168
-  },
-  "weight_map": {
-    "lm_head.weight": "model-00001-of-00002.safetensors",
-    "model.image_newline": "model-00001-of-00002.safetensors",
-    "model.language_model.embed_tokens.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.norm.weight": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_1.bias": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_1.weight": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_2.bias": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.final_layernorm.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.final_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.patch_embed.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.patch_embed.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.pos_embed.weight": "model-00002-of-00002.safetensors"
-  }
-}
diff --git a/ckpts/4b_v9_1/modeling_qwen3_5vit_qwen3.py b/ckpts/4b_v9_1/modeling_qwen3_5vit_qwen3.py
deleted file mode 100644
index b3f5b907b874422828e6550e4d0e9393d572a96c..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/modeling_qwen3_5vit_qwen3.py
+++ /dev/null
@@ -1,351 +0,0 @@
-"""
-modeling_qwen3_5vit_qwen3.py — Qwen3.5 Vision as SigLIP-compat vision_tower in LlavaOnevision.
-
-设计：与 `modeling_qwen3vlvit_qwen3.py` 严格同构，仅换 vision backbone 源：
-  - Qwen3VLVisionModel → Qwen3_5VisionModel（继承关系：Qwen3_5VisionModel(Qwen3VLVisionModel) 去 DeepStack）
-  - Qwen3VLVisionConfig → Qwen3_5VisionConfig（父类用 AttributeError 哨兵屏蔽 deepstack_visual_indexes）
-
-其余（Adapter 契约翻译、MLP projector + pre_norm、LlavaOnevision 继承 wire class）与
-Qwen3-VL ViT pipeline 完全一致。两条 pipeline 并存意义：DeepStack ablation 天然实验组。
-
-类层级：
-  Qwen3_5ViTBackbone(Qwen3_5VisionModel)      — 去 merger，保持 NaViT 契约
-  Qwen3_5ViTAsSiglipAdapter(nn.Module)        — 持有 Backbone，做 SigLIP ↔ NaViT 契约翻译
-
-三方对比公平性：定 384×384 AnyRes tile + 同款 projector 骨架 + 同款 Qwen3-1.7B LLM。
-"""
-
-import math
-import os
-import sys
-from typing import Optional
-
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-from transformers import (
-    AutoConfig,
-    AutoModel,
-    AutoModelForCausalLM,
-    LlavaOnevisionConfig,
-    LlavaOnevisionForConditionalGeneration,
-    LlavaOnevisionModel,
-    LlavaOnevisionPreTrainedModel,
-    Qwen3Config,
-)
-from transformers.activations import ACT2FN
-from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
-from transformers.models.qwen3_5.configuration_qwen3_5 import Qwen3_5VisionConfig
-from transformers.models.qwen3_5.modeling_qwen3_5 import Qwen3_5VisionModel
-
-# Shared layout-permutation utility lives in declip_qwenvit (single source of
-# truth — same code path runs in declip-training-side qk_cosine reorder).
-# Add VisionEncoder repo root to sys.path so this modeling file is importable
-# even when the package isn't pip-installed (ms-swift integration loads it
-# via dynamic plugin path).
-_REPO_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
-if _REPO_ROOT not in sys.path:
-    sys.path.append(_REPO_ROOT)
-from declip_qwenvit.model.qwen3vit_qk import block_merge_to_row_major_permutation  # noqa: E402
-
-
-class LlavaQwen3_5ViTConfig(LlavaOnevisionConfig):
-    """LlavaOnevisionConfig 子类，vision_config 类型换成 Qwen3_5VisionConfig。
-
-    与 Qwen3-VL ViT 版的差异：
-      - sub_configs.vision_config 用 Qwen3_5VisionConfig
-      - 不再设置 deepstack_visual_indexes（Qwen3_5VisionConfig 用 AttributeError 哨兵屏蔽此字段）
-
-    其余同 LlavaQwen3VLViTConfig（tile_size 默认 384；vision_feature_select_strategy='full'
-    必须 override，Qwen3.5 ViT 无 CLS token）。
-    """
-
-    model_type = "llava_qwen3_5vit_qwen3"
-    sub_configs = {"vision_config": Qwen3_5VisionConfig, "text_config": Qwen3Config}
-
-    def __init__(
-        self,
-        vision_config=None,
-        text_config=None,
-        tile_size: int = 384,
-        **kwargs,
-    ):
-        if isinstance(vision_config, dict):
-            vision_config = Qwen3_5VisionConfig(**vision_config)
-        elif vision_config is None:
-            vision_config = Qwen3_5VisionConfig()
-        # WHY 无 `vision_config.deepstack_visual_indexes = []`（对比 Qwen3-VL ViT 版）：
-        # Qwen3_5VisionConfig 父类用 AttributeError() 哨兵显式屏蔽此字段，设置会报 AttributeError
-        # LlavaOnevision.pack_image_features 用这个作为 tile 像素大小（不是 patch_size）
-        vision_config.image_size = tile_size
-
-        if isinstance(text_config, dict):
-            text_config = Qwen3Config(**text_config)
-        elif text_config is None:
-            text_config = Qwen3Config()
-
-        # 父类默认 select_strategy='default' 会跳首 token (CLS) — Qwen3.5 ViT 无 CLS 必须用 'full'
-        kwargs.setdefault("vision_feature_select_strategy", "full")
-        super().__init__(vision_config=vision_config, text_config=text_config, **kwargs)
-
-
-class Qwen3_5ViTBackbone(Qwen3_5VisionModel):
-    """Qwen3.5 Vision 去除原生 patch_merger 的 backbone 版本（V6 final_layernorm fix (2026-05-16): append final_layernorm）。
-
-    构造时把 merger.norm 的预训练权重抠到 final_layernorm，然后 `del self.merger`
-    释放 ~37M 参数（保留 norm 的 LN 焊到末端做 post_layernorm 角色，丢弃 spatial
-    shuffle + linear_fc1/fc2，那对应 LlavaOV projector 的职责）。
-
-    架构对称（V6 final_layernorm fix (2026-05-16) 修复）：
-      SigLIP2:    encoder → post_layernorm → last_hidden_state → LlavaOV MLP → LLM
-      V6 final_layernorm fix (2026-05-16):     encoder → final_layernorm → last_hidden_state → LlavaOV MLP → LLM
-
-    forward 跑完 transformer blocks 后过 final_layernorm，再返回。下游 LlavaOnevision
-    pack_image_features 的 AnyRes 2×2 pool 接管原 merger 的空间合并职责。
-
-    输入输出契约与父类 Qwen3_5VisionModel 一致（NaViT flat）：
-      forward(hidden_states=[L, patch_dim], grid_thw=[N, 3])
-      → BaseModelOutput(last_hidden_state=[L, hidden_size])
-
-    forward 主体 1:1 对照 `Qwen3_5VisionModel.forward`（已无 deepstack loop，比
-    Qwen3VLVisionModel.forward 更短），仅跳过末尾 `self.merger(x)`，改为 final_layernorm。
-    """
-
-    def __init__(self, config):
-        super().__init__(config)
-        # V6 final_layernorm fix (2026-05-16): extract merger.norm pretrained weights into final_layernorm.
-        # Default init: even if ckpt 阶段没 inject final_layernorm.* (e.g. stock
-        # bootstrap path that filters merger.*), final_layernorm 仍持有 merger.norm
-        # 的预训练值, 不是 random — 这是防止 silent corruption 的兜底.
-        ln_w = self.merger.norm.weight.detach().clone()
-        ln_b = self.merger.norm.bias.detach().clone()
-        del self.merger
-        self.final_layernorm = nn.LayerNorm(config.hidden_size, eps=1e-6)
-        self.final_layernorm.weight.data.copy_(ln_w)
-        self.final_layernorm.bias.data.copy_(ln_b)
-
-    def forward(self, hidden_states, grid_thw, **kwargs):
-        hidden_states = self.patch_embed(hidden_states)
-
-        pos_embeds = self.fast_pos_embed_interpolate(grid_thw)
-        hidden_states = hidden_states + pos_embeds
-
-        rotary_pos_emb = self.rot_pos_emb(grid_thw)
-        seq_len, _ = hidden_states.size()
-        rotary_pos_emb = rotary_pos_emb.reshape(seq_len, -1)
-        emb = torch.cat((rotary_pos_emb, rotary_pos_emb), dim=-1)
-        position_embeddings = (emb.cos(), emb.sin())
-
-        cu_seqlens = torch.repeat_interleave(
-            grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]
-        ).cumsum(dim=0, dtype=torch.int32)
-        cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
-
-        for blk in self.blocks:
-            hidden_states = blk(
-                hidden_states,
-                cu_seqlens=cu_seqlens,
-                position_embeddings=position_embeddings,
-                **kwargs,
-            )
-
-        # V6 final_layernorm fix (2026-05-16): appended final LayerNorm — mirrors SigLIP2's post_layernorm.
-        # Per-token affine; layout-invariant (reorder happens in adapter).
-        hidden_states = self.final_layernorm(hidden_states)
-
-        return BaseModelOutput(last_hidden_state=hidden_states)
-
-
-class Qwen3_5ViTAsSiglipAdapter(nn.Module):
-    """SigLIP 契约 → NaViT 契约的翻译层。持有 Qwen3_5ViTBackbone。
-
-    对外暴露 SigLIP 式 forward(pixel_values=[N,3,H,W]) → BaseModelOutputWithPooling，
-    供 LlavaOnevision 消费；对内按官方 _preprocess 的 reshape 链把 pixel_values
-    转成 NaViT flat + grid_thw 喂给 Backbone。
-
-    reshape 链 1:1 照抄 transformers 官方 Qwen2VLImageProcessorFast._preprocess
-    (video_processing_qwen3_vl.py L227-252) —— Qwen3.5 无独立 image_processor，复用 Qwen3-VL 格式。
-    """
-
-    def __init__(self, vision_config: Qwen3_5VisionConfig):
-        super().__init__()
-        self.vision = Qwen3_5ViTBackbone(vision_config)
-        self.config = vision_config
-
-    @property
-    def dtype(self):
-        return next(self.parameters()).dtype
-
-    @property
-    def device(self):
-        return next(self.parameters()).device
-
-    def _flatten_navit(self, pixel_values: torch.Tensor):
-        """[N, 3, H, W] → (flat=[N*L, patch_dim], grid_thw=[N, 3], shape=(N, L)).
-
-        L = grid_t * grid_h * grid_w = 1 * (H/16) * (W/16)
-        patch_dim = C * temporal_patch_size * patch_size^2 = 3 * 2 * 16 * 16 = 1536
-        """
-        pixel_values = pixel_values.to(dtype=self.dtype)
-        tps = self.config.temporal_patch_size
-        ps = self.config.patch_size
-        ms = self.config.spatial_merge_size
-
-        patches = pixel_values.unsqueeze(1)
-        # 对单帧图像 T=1, pad=1 → expand 一帧使 T 整除 temporal_patch_size，
-        # Conv3d 在复制帧上退化为等效 2D Conv（数学无损）
-        T = patches.shape[1]
-        pad = -T % tps
-        if pad:
-            repeats = patches[:, -1:].expand(-1, pad, -1, -1, -1)
-            patches = torch.cat((patches, repeats), dim=1)
-
-        batch_size, t, channel, H, W = patches.shape
-        grid_t = t // tps
-        grid_h = H // ps
-        grid_w = W // ps
-
-        patches = patches.view(
-            batch_size, grid_t, tps, channel,
-            grid_h // ms, ms, ps,
-            grid_w // ms, ms, ps,
-        )
-        patches = patches.permute(0, 1, 4, 7, 5, 8, 3, 2, 6, 9)
-        flatten_patches = patches.reshape(
-            batch_size,
-            grid_t * grid_h * grid_w,
-            channel * tps * ps * ps,
-        )
-
-        seq_len = grid_t * grid_h * grid_w
-        flat = flatten_patches.reshape(batch_size * seq_len, -1)
-        # on-device 构造小 tensor 再 expand，host→GPU 同步量 O(3) 而非 O(N*3)
-        grid_unit = torch.tensor(
-            [grid_t, grid_h, grid_w], dtype=torch.int32, device=pixel_values.device,
-        )
-        grid_thw = grid_unit.unsqueeze(0).expand(batch_size, -1).contiguous()
-        return flat, grid_thw, (batch_size, seq_len)
-
-    def forward(
-        self,
-        pixel_values: torch.Tensor,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-        **kwargs,
-    ) -> BaseModelOutputWithPooling:
-        flat, grid_thw, (N, S) = self._flatten_navit(pixel_values)
-        vision_out = self.vision(flat, grid_thw=grid_thw)
-        hidden = vision_out.last_hidden_state.view(N, S, -1)
-
-        # Block-merge → row-major reorder before handing to LlavaOnevision.
-        # Internally the ViT runs in Qwen NaViT block-merge layout (pretrained
-        # pos_embed + RoPE contract); downstream LlavaOV `pack_image_features`
-        # (multi-tile AnyRes path, view(num_patch_h, num_patch_w, h, w, -1))
-        # and `apply_pooling` (video path, view(B, h, w, -1) + bilinear) BOTH
-        # assume row-major. Without this reorder, the multi-tile/video spatial
-        # pool pulls together tokens that are NOT spatially adjacent — silent
-        # corruption that doesn't fire on S1 single-tile path (line 348-351 of
-        # modeling_llava_onevision.py just flattens [N,D] verbatim) but kills
-        # S2 / eval quality.
-        grid_h = int(grid_thw[0, 1].item())
-        grid_w = int(grid_thw[0, 2].item())
-        ms = getattr(self.config, "spatial_merge_size", 2)
-        layout_perm = block_merge_to_row_major_permutation(
-            grid_h, grid_w, ms=ms, device=hidden.device,
-        )
-        hidden = hidden[:, layout_perm, :]
-
-        return BaseModelOutputWithPooling(
-            last_hidden_state=hidden,
-            # LlavaOnevision 索引 hidden_states[vision_feature_layer=-1]；tuple 长度 1 足够
-            hidden_states=(hidden,),
-            pooler_output=None,
-        )
-
-
-class LlavaQwen3_5ViTMultiModalProjector(nn.Module):
-    """标准 LlavaOnevision projector（V6 final_layernorm fix (2026-05-16): pre_norm → Identity）。
-
-    V6 final_layernorm fix (2026-05-16) 修复后, encoder 末端已自带 final_layernorm（与 SigLIP2 post_layernorm 对称），
-    projector 不再需要补 LN — pre_norm 改为 nn.Identity，对齐 SigLIP2 plugin 的
-    LlavaOnevision stock projector 结构（裸 linear_1 → GELU → linear_2），
-    保证 SigLIP2 / Qwen3.5 / Qwen3-VL 三个 backbone 在 LlavaOV 设定下公平对比。
-
-    （历史：V6.0.0~V6.0.4 时期 encoder 无 final LN，projector pre_norm 是补丁；
-    现在补丁回到 encoder 内部，projector 回归 stock 形态。）
-    """
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        super().__init__()
-        num_feature_layers = (
-            1 if isinstance(config.vision_feature_layer, int) else len(config.vision_feature_layer)
-        )
-        vision_dim = config.vision_config.hidden_size * num_feature_layers
-        text_dim = config.text_config.hidden_size
-        bias = getattr(config, "multimodal_projector_bias", True)
-
-        # V6 final_layernorm fix (2026-05-16): pre_norm = Identity (encoder 已自带 final_layernorm).
-        self.pre_norm = nn.Identity()
-        self.linear_1 = nn.Linear(vision_dim, text_dim, bias=bias)
-        self.act = ACT2FN[config.projector_hidden_act]
-        self.linear_2 = nn.Linear(text_dim, text_dim, bias=bias)
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        return self.linear_2(self.act(self.linear_1(self.pre_norm(x))))
-
-
-class LlavaQwen3_5ViTModel(LlavaOnevisionModel):
-    """继承 LlavaOnevisionModel 但绕过其 __init__ 手动装配。
-
-    父类 __init__ 调 `AutoModel.from_config(config.vision_config)` 会对 Qwen3_5VisionConfig
-    抛 "Unrecognized configuration"（Qwen3.5 vision 没注册到 AutoModel）。手动装配避开
-    这一步，同时省掉"先构造 Qwen3_5VisionModel 再被替换"的双重开销（~1.3GB init-peak）。
-
-    装配顺序与父类一致：vision_tower / projector / image_newline / language_model / post_init。
-    """
-
-    config_class = LlavaQwen3_5ViTConfig
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        # 跳过 LlavaOnevisionModel.__init__（AutoModel 不识别 Qwen3_5VisionConfig）
-        LlavaOnevisionPreTrainedModel.__init__(self, config)
-        self.vision_tower = Qwen3_5ViTAsSiglipAdapter(config.vision_config)
-        self.multi_modal_projector = LlavaQwen3_5ViTMultiModalProjector(config)
-        embed_std = 1 / math.sqrt(config.text_config.hidden_size)
-        self.image_newline = nn.Parameter(
-            torch.randn(config.text_config.hidden_size, dtype=self.dtype) * embed_std
-        )
-        self.vocab_size = config.text_config.vocab_size
-        self.language_model = AutoModel.from_config(config.text_config)
-        self.post_init()
-
-
-class LlavaQwen3_5ViTForConditionalGeneration(LlavaOnevisionForConditionalGeneration):
-    """继承 LlavaOnevisionForConditionalGeneration，只换 self.model。
-
-    同样跳过父类 __init__（避免重复构造 LlavaOnevisionModel，根因见 LlavaQwen3_5ViTModel）。
-    """
-
-    config_class = LlavaQwen3_5ViTConfig
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        LlavaOnevisionPreTrainedModel.__init__(self, config)
-        self.model = LlavaQwen3_5ViTModel(config)
-        self.lm_head = nn.Linear(
-            config.text_config.hidden_size, config.text_config.vocab_size, bias=False
-        )
-        self.post_init()
-
-
-AutoConfig.register(LlavaQwen3_5ViTConfig.model_type, LlavaQwen3_5ViTConfig)
-AutoModelForCausalLM.register(LlavaQwen3_5ViTConfig, LlavaQwen3_5ViTForConditionalGeneration)
-
-
-__all__ = [
-    "LlavaQwen3_5ViTConfig",
-    "Qwen3_5ViTBackbone",
-    "Qwen3_5ViTAsSiglipAdapter",
-    "LlavaQwen3_5ViTMultiModalProjector",
-    "LlavaQwen3_5ViTModel",
-    "LlavaQwen3_5ViTForConditionalGeneration",
-]
diff --git a/ckpts/4b_v9_1/processor_config.json b/ckpts/4b_v9_1/processor_config.json
deleted file mode 100644
index 204ed5d216c648cbd71da6c25d67b4bf1ec84433..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/processor_config.json
+++ /dev/null
@@ -1,203 +0,0 @@
-{
-  "image_processor": {
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_pad": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "image_grid_pinpoints": [
-      [
-        384,
-        384
-      ],
-      [
-        384,
-        768
-      ],
-      [
-        384,
-        1152
-      ],
-      [
-        384,
-        1536
-      ],
-      [
-        384,
-        1920
-      ],
-      [
-        384,
-        2304
-      ],
-      [
-        768,
-        384
-      ],
-      [
-        768,
-        768
-      ],
-      [
-        768,
-        1152
-      ],
-      [
-        768,
-        1536
-      ],
-      [
-        768,
-        1920
-      ],
-      [
-        768,
-        2304
-      ],
-      [
-        1152,
-        384
-      ],
-      [
-        1152,
-        768
-      ],
-      [
-        1152,
-        1152
-      ],
-      [
-        1152,
-        1536
-      ],
-      [
-        1152,
-        1920
-      ],
-      [
-        1152,
-        2304
-      ],
-      [
-        1536,
-        384
-      ],
-      [
-        1536,
-        768
-      ],
-      [
-        1536,
-        1152
-      ],
-      [
-        1536,
-        1536
-      ],
-      [
-        1536,
-        1920
-      ],
-      [
-        1536,
-        2304
-      ],
-      [
-        1920,
-        384
-      ],
-      [
-        1920,
-        768
-      ],
-      [
-        1920,
-        1152
-      ],
-      [
-        1920,
-        1536
-      ],
-      [
-        1920,
-        1920
-      ],
-      [
-        1920,
-        2304
-      ],
-      [
-        2304,
-        384
-      ],
-      [
-        2304,
-        768
-      ],
-      [
-        2304,
-        1152
-      ],
-      [
-        2304,
-        1536
-      ],
-      [
-        2304,
-        1920
-      ],
-      [
-        2304,
-        2304
-      ]
-    ],
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_processor_type": "LlavaOnevisionImageProcessor",
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "size": {
-      "height": 384,
-      "width": 384
-    }
-  },
-  "image_token": "<image>",
-  "num_image_tokens": 576,
-  "processor_class": "LlavaOnevisionProcessor",
-  "video_processor": {
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "do_sample_frames": false,
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "return_metadata": false,
-    "size": {
-      "height": 384,
-      "width": 384
-    },
-    "video_processor_type": "LlavaOnevisionVideoProcessor"
-  },
-  "video_token": "<video>",
-  "vision_aspect_ratio": "anyres_max_9",
-  "vision_feature_select_strategy": null
-}
diff --git a/ckpts/4b_v9_1/tokenizer.json b/ckpts/4b_v9_1/tokenizer.json
deleted file mode 100644
index c40d89c9226d696c215d65a68eea68c6dd55d79e..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/tokenizer.json
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:9024318c850eaadf26be79389d21b07a7afd8f1af749b89f9109b06c0d12173c
-size 11423018
diff --git a/ckpts/4b_v9_1/tokenizer_config.json b/ckpts/4b_v9_1/tokenizer_config.json
deleted file mode 100644
index 91a90671a687fb5d1fc3ed576d216e5e67d9e11b..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/tokenizer_config.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "add_prefix_space": false,
-  "backend": "tokenizers",
-  "bos_token": null,
-  "clean_up_tokenization_spaces": false,
-  "eos_token": "<|im_end|>",
-  "errors": "replace",
-  "extra_special_tokens": [
-    "<image>",
-    "<video>"
-  ],
-  "is_local": true,
-  "model_max_length": 1010000,
-  "pad_token": "<|endoftext|>",
-  "processor_class": "LlavaOnevisionProcessor",
-  "split_special_tokens": false,
-  "tokenizer_class": "Qwen2Tokenizer",
-  "unk_token": null
-}
diff --git a/ckpts/4b_v9_1/zero_to_fp32.py b/ckpts/4b_v9_1/zero_to_fp32.py
deleted file mode 100644
index 5995d6e6f04e43b989587aa9022a3aef0c66d694..0000000000000000000000000000000000000000
--- a/ckpts/4b_v9_1/zero_to_fp32.py
+++ /dev/null
@@ -1,760 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright (c) Microsoft Corporation.
-# SPDX-License-Identifier: Apache-2.0
-
-# DeepSpeed Team
-
-# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
-# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
-# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
-# application.
-#
-# example:
-#   python zero_to_fp32.py . output_dir/
-#   or
-#   python zero_to_fp32.py . output_dir/ --safe_serialization
-
-import argparse
-import torch
-import glob
-import math
-import os
-import re
-import gc
-import json
-import numpy as np
-from tqdm import tqdm
-from collections import OrderedDict
-from dataclasses import dataclass
-
-# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
-# DeepSpeed data structures it has to be available in the current python environment.
-from deepspeed.utils import logger
-from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
-                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
-                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
-
-
-@dataclass
-class zero_model_state:
-    buffers: dict()
-    param_shapes: dict()
-    shared_params: list
-    ds_version: int
-    frozen_param_shapes: dict()
-    frozen_param_fragments: dict()
-
-
-debug = 0
-
-# load to cpu
-device = torch.device('cpu')
-
-
-def atoi(text):
-    return int(text) if text.isdigit() else text
-
-
-def natural_keys(text):
-    '''
-    alist.sort(key=natural_keys) sorts in human order
-    http://nedbatchelder.com/blog/200712/human_sorting.html
-    (See Toothy's implementation in the comments)
-    '''
-    return [atoi(c) for c in re.split(r'(\d+)', text)]
-
-
-def get_model_state_file(checkpoint_dir, zero_stage):
-    if not os.path.isdir(checkpoint_dir):
-        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
-
-    # there should be only one file
-    if zero_stage <= 2:
-        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
-    elif zero_stage == 3:
-        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
-
-    if not os.path.exists(file):
-        raise FileNotFoundError(f"can't find model states file at '{file}'")
-
-    return file
-
-
-def get_checkpoint_files(checkpoint_dir, glob_pattern):
-    # XXX: need to test that this simple glob rule works for multi-node setup too
-    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
-
-    if len(ckpt_files) == 0:
-        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
-
-    return ckpt_files
-
-
-def get_optim_files(checkpoint_dir):
-    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
-
-
-def get_model_state_files(checkpoint_dir):
-    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
-
-
-def parse_model_states(files):
-    zero_model_states = []
-    for file in files:
-        state_dict = torch.load(file, map_location=device, weights_only=False)
-
-        if BUFFER_NAMES not in state_dict:
-            raise ValueError(f"{file} is not a model state checkpoint")
-        buffer_names = state_dict[BUFFER_NAMES]
-        if debug:
-            print("Found buffers:", buffer_names)
-
-        # recover just the buffers while restoring them to fp32 if they were saved in fp16
-        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
-        param_shapes = state_dict[PARAM_SHAPES]
-
-        # collect parameters that are included in param_shapes
-        param_names = []
-        for s in param_shapes:
-            for name in s.keys():
-                param_names.append(name)
-
-        # update with frozen parameters
-        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
-        if frozen_param_shapes is not None:
-            if debug:
-                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
-            param_names += list(frozen_param_shapes.keys())
-
-        # handle shared params
-        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
-
-        ds_version = state_dict.get(DS_VERSION, None)
-
-        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
-
-        z_model_state = zero_model_state(buffers=buffers,
-                                         param_shapes=param_shapes,
-                                         shared_params=shared_params,
-                                         ds_version=ds_version,
-                                         frozen_param_shapes=frozen_param_shapes,
-                                         frozen_param_fragments=frozen_param_fragments)
-        zero_model_states.append(z_model_state)
-
-    return zero_model_states
-
-
-def parse_optim_states(files, ds_checkpoint_dir):
-    total_files = len(files)
-    state_dicts = []
-    for f in tqdm(files, desc='Loading checkpoint shards'):
-        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
-        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
-        # and also handle the case where it was already removed by another helper script
-        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
-        state_dicts.append(state_dict)
-
-    if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
-        raise ValueError(f"{files[0]} is not a zero checkpoint")
-    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
-    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
-
-    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
-    # parameters can be different from data parallelism for non-expert parameters. So we can just
-    # use the max of the partition_count to get the dp world_size.
-
-    if type(world_size) is list:
-        world_size = max(world_size)
-
-    if world_size != total_files:
-        raise ValueError(
-            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
-            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
-        )
-
-    # the groups are named differently in each stage
-    if zero_stage <= 2:
-        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
-    elif zero_stage == 3:
-        fp32_groups_key = FP32_FLAT_GROUPS
-    else:
-        raise ValueError(f"unknown zero stage {zero_stage}")
-
-    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
-    return zero_stage, world_size, fp32_flat_groups
-
-
-def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
-    """
-    Returns fp32 state_dict reconstructed from ds checkpoint
-
-    Args:
-        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
-
-    """
-    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
-
-    optim_files = get_optim_files(ds_checkpoint_dir)
-    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
-    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
-
-    model_files = get_model_state_files(ds_checkpoint_dir)
-
-    zero_model_states = parse_model_states(model_files)
-    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
-
-    if zero_stage <= 2:
-        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                                          exclude_frozen_parameters)
-    elif zero_stage == 3:
-        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                                          exclude_frozen_parameters)
-
-
-def _zero2_merge_frozen_params(state_dict, zero_model_states):
-    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
-        return
-
-    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
-    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
-
-    if debug:
-        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
-        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
-
-        wanted_params = len(frozen_param_shapes)
-        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
-        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
-        print(f'Frozen params: Have {avail_numel} numels to process.')
-        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
-
-    total_params = 0
-    total_numel = 0
-    for name, shape in frozen_param_shapes.items():
-        total_params += 1
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-
-        state_dict[name] = frozen_param_fragments[name]
-
-        if debug:
-            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
-
-    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _has_callable(obj, fn):
-    attr = getattr(obj, fn, None)
-    return callable(attr)
-
-
-def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
-    param_shapes = zero_model_states[0].param_shapes
-
-    # Reconstruction protocol:
-    #
-    # XXX: document this
-
-    if debug:
-        for i in range(world_size):
-            for j in range(len(fp32_flat_groups[0])):
-                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
-
-    # XXX: memory usage doubles here (zero2)
-    num_param_groups = len(fp32_flat_groups[0])
-    merged_single_partition_of_fp32_groups = []
-    for i in range(num_param_groups):
-        merged_partitions = [sd[i] for sd in fp32_flat_groups]
-        full_single_fp32_vector = torch.cat(merged_partitions, 0)
-        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
-    avail_numel = sum(
-        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
-
-    if debug:
-        wanted_params = sum([len(shapes) for shapes in param_shapes])
-        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
-        # not asserting if there is a mismatch due to possible padding
-        print(f"Have {avail_numel} numels to process.")
-        print(f"Need {wanted_numel} numels in {wanted_params} params.")
-
-    # params
-    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
-    # out-of-core computing solution
-    total_numel = 0
-    total_params = 0
-    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
-        offset = 0
-        avail_numel = full_single_fp32_vector.numel()
-        for name, shape in shapes.items():
-
-            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
-            total_numel += unpartitioned_numel
-            total_params += 1
-
-            if debug:
-                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
-            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
-            offset += unpartitioned_numel
-
-        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
-        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
-        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
-        # live optimizer object, so we are checking that the numbers are within the right range
-        align_to = 2 * world_size
-
-        def zero2_align(x):
-            return align_to * math.ceil(x / align_to)
-
-        if debug:
-            print(f"original offset={offset}, avail_numel={avail_numel}")
-
-        offset = zero2_align(offset)
-        avail_numel = zero2_align(avail_numel)
-
-        if debug:
-            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
-
-        # Sanity check
-        if offset != avail_numel:
-            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
-
-    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                               exclude_frozen_parameters):
-    state_dict = OrderedDict()
-
-    # buffers
-    buffers = zero_model_states[0].buffers
-    state_dict.update(buffers)
-    if debug:
-        print(f"added {len(buffers)} buffers")
-
-    if not exclude_frozen_parameters:
-        _zero2_merge_frozen_params(state_dict, zero_model_states)
-
-    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
-
-    # recover shared parameters
-    for pair in zero_model_states[0].shared_params:
-        if pair[1] in state_dict:
-            state_dict[pair[0]] = state_dict[pair[1]]
-
-    return state_dict
-
-
-def zero3_partitioned_param_info(unpartitioned_numel, world_size):
-    remainder = unpartitioned_numel % world_size
-    padding_numel = (world_size - remainder) if remainder else 0
-    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
-    return partitioned_numel, padding_numel
-
-
-def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
-    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
-        return
-
-    if debug:
-        for i in range(world_size):
-            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
-            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
-
-        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
-        wanted_params = len(frozen_param_shapes)
-        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
-        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
-        print(f'Frozen params: Have {avail_numel} numels to process.')
-        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
-
-    total_params = 0
-    total_numel = 0
-    for name, shape in zero_model_states[0].frozen_param_shapes.items():
-        total_params += 1
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-
-        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
-        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
-
-        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
-
-        if debug:
-            print(
-                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
-            )
-
-    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
-
-
-class GatheredTensor:
-    """
-    A pseudo tensor that collects partitioned weights.
-    It is more memory efficient when there are multiple groups.
-    """
-
-    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
-        self.flat_groups = flat_groups
-        self.flat_groups_offset = flat_groups_offset
-        self.offset = offset
-        self.partitioned_numel = partitioned_numel
-        self.shape = shape
-        self.dtype = self.flat_groups[0][0].dtype
-
-    def contiguous(self):
-        """
-        Merge partitioned weights from flat_groups into a single tensor.
-        """
-        end_idx = self.offset + self.partitioned_numel
-        world_size = len(self.flat_groups)
-        pad_flat_param_chunks = []
-
-        for rank_i in range(world_size):
-            # for each rank, we need to collect weights from related group/groups
-            flat_groups_at_rank_i = self.flat_groups[rank_i]
-            start_group_id = None
-            end_group_id = None
-            for group_id in range(len(self.flat_groups_offset)):
-                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
-                    start_group_id = group_id
-                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
-                    end_group_id = group_id
-                    break
-            # collect weights from related group/groups
-            for group_id in range(start_group_id, end_group_id + 1):
-                flat_tensor = flat_groups_at_rank_i[group_id]
-                start_offset = self.offset - self.flat_groups_offset[group_id]
-                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
-                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
-
-        # collect weights from all ranks
-        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
-        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
-        return param
-
-
-def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
-    param_shapes = zero_model_states[0].param_shapes
-    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
-
-    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
-    # param, re-consolidating each param, while dealing with padding if any
-
-    # merge list of dicts, preserving order
-    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
-
-    if debug:
-        for i in range(world_size):
-            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
-
-        wanted_params = len(param_shapes)
-        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
-        # not asserting if there is a mismatch due to possible padding
-        avail_numel = fp32_flat_groups[0].numel() * world_size
-        print(f"Trainable params: Have {avail_numel} numels to process.")
-        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
-
-    # params
-    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
-    # out-of-core computing solution
-    offset = 0
-    total_numel = 0
-    total_params = 0
-    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
-    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-        total_params += 1
-        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
-
-        if debug:
-            print(
-                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
-            )
-
-        # memory efficient tensor
-        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
-        state_dict[name] = tensor
-        offset += partitioned_numel
-
-    offset *= world_size
-
-    # Sanity check
-    if offset != avail_numel:
-        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
-
-    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                               exclude_frozen_parameters):
-    state_dict = OrderedDict()
-
-    # buffers
-    buffers = zero_model_states[0].buffers
-    state_dict.update(buffers)
-    if debug:
-        print(f"added {len(buffers)} buffers")
-
-    if not exclude_frozen_parameters:
-        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
-
-    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
-
-    # recover shared parameters
-    for pair in zero_model_states[0].shared_params:
-        if pair[1] in state_dict:
-            state_dict[pair[0]] = state_dict[pair[1]]
-
-    return state_dict
-
-
-def to_torch_tensor(state_dict, return_empty_tensor=False):
-    """
-    Convert state_dict of GatheredTensor to torch tensor
-    """
-    torch_state_dict = {}
-    converted_tensors = {}
-    for name, tensor in state_dict.items():
-        tensor_id = id(tensor)
-        if tensor_id in converted_tensors:  # shared tensors
-            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
-            torch_state_dict[name] = shared_tensor
-        else:
-            converted_tensors[tensor_id] = name
-            if return_empty_tensor:
-                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
-            else:
-                torch_state_dict[name] = tensor.contiguous()
-    return torch_state_dict
-
-
-def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
-                                             tag=None,
-                                             exclude_frozen_parameters=False,
-                                             lazy_mode=False):
-    """
-    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
-    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
-    via a model hub.
-
-    Args:
-        - ``checkpoint_dir``: path to the desired checkpoint folder
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
-        - ``exclude_frozen_parameters``: exclude frozen parameters
-        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
-          Convert the pesduo tensor to torch tensor by ``.contiguous()``
-
-    Returns:
-        - pytorch ``state_dict``
-
-    A typical usage might be ::
-
-        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
-        # do the training and checkpoint saving
-        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
-        model = model.cpu() # move to cpu
-        model.load_state_dict(state_dict)
-        # submit to model hub or save the model to share with others
-
-    In this example the ``model`` will no longer be usable in the deepspeed context of the same
-    application. i.e. you will need to re-initialize the deepspeed engine, since
-    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
-
-    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
-
-    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
-    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
-    the checkpoint. Or you can load state_dict in lazy mode ::
-
-        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
-        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
-        for name, lazy_tensor in state_dict.item():
-            tensor = lazy_tensor.contiguous()  # to cpu
-            print(name, tensor)
-            # del tensor to release memory if it no longer in use
-    """
-    if tag is None:
-        latest_path = os.path.join(checkpoint_dir, 'latest')
-        if os.path.isfile(latest_path):
-            with open(latest_path, 'r') as fd:
-                tag = fd.read().strip()
-        else:
-            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
-
-    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
-
-    if not os.path.isdir(ds_checkpoint_dir):
-        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
-
-    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
-    if lazy_mode:
-        return state_dict
-    else:
-        return to_torch_tensor(state_dict)
-
-
-def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
-                                               output_dir,
-                                               max_shard_size="5GB",
-                                               safe_serialization=False,
-                                               tag=None,
-                                               exclude_frozen_parameters=False):
-    """
-    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
-    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
-
-    Args:
-        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
-        - ``output_dir``: directory to the pytorch fp32 state_dict output files
-        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
-        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
-        - ``exclude_frozen_parameters``: exclude frozen parameters
-    """
-
-    # Dependency pre-check
-    if safe_serialization:
-        try:
-            from safetensors.torch import save_file
-        except ImportError:
-            print('If you want to use `safe_serialization`, please `pip install safetensors`')
-            raise
-    if max_shard_size is not None:
-        try:
-            from huggingface_hub import split_torch_state_dict_into_shards
-        except ImportError:
-            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
-            raise
-
-    # Convert zero checkpoint to state_dict
-    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
-                                                          tag,
-                                                          exclude_frozen_parameters,
-                                                          lazy_mode=True)
-
-    # Shard the model if it is too big.
-    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
-    if max_shard_size is not None:
-        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
-        # an memory-efficient approach for sharding
-        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
-        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
-                                                              filename_pattern=filename_pattern,
-                                                              max_shard_size=max_shard_size)
-    else:
-        from collections import namedtuple
-        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
-        state_dict_split = StateDictSplit(is_sharded=False,
-                                          filename_to_tensors={weights_name: list(state_dict.keys())})
-
-    # Save the model by shard
-    os.makedirs(output_dir, exist_ok=True)
-    filename_to_tensors = state_dict_split.filename_to_tensors.items()
-    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
-        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
-        shard_state_dict = to_torch_tensor(shard_state_dict)
-        output_path = os.path.join(output_dir, shard_file)
-        if safe_serialization:
-            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
-        else:
-            torch.save(shard_state_dict, output_path)
-        # release the memory of current shard
-        for tensor_name in list(shard_state_dict.keys()):
-            del state_dict[tensor_name]
-            del shard_state_dict[tensor_name]
-        del shard_state_dict
-        gc.collect()
-
-    # Save index if sharded
-    if state_dict_split.is_sharded:
-        index = {
-            "metadata": state_dict_split.metadata,
-            "weight_map": state_dict_split.tensor_to_filename,
-        }
-        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
-        save_index_file = os.path.join(output_dir, save_index_file)
-        with open(save_index_file, "w", encoding="utf-8") as f:
-            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
-            f.write(content)
-
-
-def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
-    """
-    1. Put the provided model to cpu
-    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
-    3. Load it into the provided model
-
-    Args:
-        - ``model``: the model object to update
-        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
-
-    Returns:
-        - ``model`: modified model
-
-    Make sure you have plenty of CPU memory available before you call this function. If you don't
-    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
-    conveniently placed for you in the checkpoint folder.
-
-    A typical usage might be ::
-
-        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
-        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
-        # submit to model hub or save the model to share with others
-
-    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
-    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
-    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
-
-    """
-    logger.info("Extracting fp32 weights")
-    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
-
-    logger.info("Overwriting model with fp32 weights")
-    model = model.cpu()
-    model.load_state_dict(state_dict, strict=False)
-
-    return model
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument("checkpoint_dir",
-                        type=str,
-                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
-    parser.add_argument("output_dir",
-                        type=str,
-                        help="directory to the pytorch fp32 state_dict output files"
-                        "(e.g. path/checkpoint-12-output/)")
-    parser.add_argument(
-        "--max_shard_size",
-        type=str,
-        default="5GB",
-        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
-        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
-        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
-        "without CPU OOM issues.")
-    parser.add_argument(
-        "--safe_serialization",
-        default=False,
-        action='store_true',
-        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
-    parser.add_argument("-t",
-                        "--tag",
-                        type=str,
-                        default=None,
-                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
-    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
-    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
-    args = parser.parse_args()
-
-    debug = args.debug
-
-    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
-                                               args.output_dir,
-                                               max_shard_size=args.max_shard_size,
-                                               safe_serialization=args.safe_serialization,
-                                               tag=args.tag,
-                                               exclude_frozen_parameters=args.exclude_frozen_parameters)
diff --git a/ckpts/stock_32f/args.json b/ckpts/stock_32f/args.json
deleted file mode 100644
index 14718b6fe3713a117336865a11f1279580575c02..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/args.json
+++ /dev/null
@@ -1,376 +0,0 @@
-{
-  "output_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_stock_32f_10pct/v0-20260606-044138",
-  "per_device_train_batch_size": 1,
-  "num_train_epochs": 1.0,
-  "max_steps": -1,
-  "learning_rate": 1e-05,
-  "lr_scheduler_type": "cosine",
-  "lr_scheduler_kwargs": null,
-  "warmup_steps": 0,
-  "optim": "adamw_torch_fused",
-  "optim_args": null,
-  "weight_decay": 0.1,
-  "adam_beta1": 0.9,
-  "adam_beta2": 0.95,
-  "adam_epsilon": 1e-08,
-  "optim_target_modules": null,
-  "gradient_accumulation_steps": 8,
-  "average_tokens_across_devices": true,
-  "max_grad_norm": 1.0,
-  "label_smoothing_factor": 0.0,
-  "bf16": true,
-  "fp16": false,
-  "bf16_full_eval": false,
-  "fp16_full_eval": false,
-  "tf32": null,
-  "gradient_checkpointing": true,
-  "gradient_checkpointing_kwargs": "{\"use_reentrant\": false}",
-  "torch_compile": false,
-  "torch_compile_backend": null,
-  "torch_compile_mode": null,
-  "use_liger_kernel": false,
-  "liger_kernel_config": null,
-  "use_cache": false,
-  "neftune_noise_alpha": null,
-  "torch_empty_cache_steps": null,
-  "auto_find_batch_size": false,
-  "logging_strategy": "steps",
-  "logging_steps": 1,
-  "logging_first_step": true,
-  "log_on_each_node": true,
-  "logging_nan_inf_filter": true,
-  "include_num_input_tokens_seen": false,
-  "log_level": "passive",
-  "log_level_replica": "warning",
-  "disable_tqdm": null,
-  "report_to": [
-    "none"
-  ],
-  "run_name": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_stock_32f_10pct/v0-20260606-044138",
-  "project": "huggingface",
-  "trackio_space_id": "trackio",
-  "eval_strategy": "no",
-  "eval_steps": 200.0,
-  "eval_delay": 0,
-  "per_device_eval_batch_size": 1,
-  "prediction_loss_only": false,
-  "eval_on_start": false,
-  "eval_do_concat_batches": true,
-  "eval_use_gather_object": false,
-  "eval_accumulation_steps": null,
-  "include_for_metrics": [],
-  "batch_eval_metrics": false,
-  "save_only_model": false,
-  "save_strategy": "steps",
-  "save_steps": 200.0,
-  "save_on_each_node": false,
-  "save_total_limit": 1,
-  "enable_jit_checkpoint": false,
-  "push_to_hub": false,
-  "hub_token": null,
-  "hub_private_repo": null,
-  "hub_model_id": null,
-  "hub_strategy": "every_save",
-  "hub_always_push": false,
-  "hub_revision": null,
-  "load_best_model_at_end": false,
-  "metric_for_best_model": "loss",
-  "greater_is_better": false,
-  "ignore_data_skip": false,
-  "restore_callback_states_from_checkpoint": false,
-  "full_determinism": false,
-  "seed": 42,
-  "data_seed": 42,
-  "use_cpu": false,
-  "accelerator_config": "{\"dispatch_batches\": false}",
-  "parallelism_config": null,
-  "dataloader_drop_last": false,
-  "dataloader_num_workers": 6,
-  "dataloader_pin_memory": true,
-  "dataloader_persistent_workers": true,
-  "dataloader_prefetch_factor": 4,
-  "remove_unused_columns": true,
-  "label_names": null,
-  "train_sampling_strategy": "random",
-  "length_column_name": "length",
-  "ddp_find_unused_parameters": null,
-  "ddp_bucket_cap_mb": null,
-  "ddp_broadcast_buffers": null,
-  "ddp_backend": null,
-  "ddp_timeout": 7200,
-  "fsdp": [],
-  "fsdp_config": null,
-  "deepspeed": {
-    "fp16": {
-      "enabled": "auto",
-      "loss_scale": 0,
-      "loss_scale_window": 1000,
-      "initial_scale_power": 16,
-      "hysteresis": 2,
-      "min_loss_scale": 1
-    },
-    "bf16": {
-      "enabled": "auto"
-    },
-    "zero_optimization": {
-      "stage": 2,
-      "offload_optimizer": {
-        "device": "none",
-        "pin_memory": true
-      },
-      "allgather_partitions": true,
-      "allgather_bucket_size": 200000000.0,
-      "overlap_comm": false,
-      "reduce_scatter": true,
-      "reduce_bucket_size": 200000000.0,
-      "contiguous_gradients": true
-    },
-    "gradient_accumulation_steps": "auto",
-    "gradient_clipping": "auto",
-    "steps_per_print": 2000,
-    "train_batch_size": "auto",
-    "train_micro_batch_size_per_gpu": "auto",
-    "wall_clock_breakdown": false
-  },
-  "debug": null,
-  "skip_memory_metrics": true,
-  "do_train": false,
-  "do_eval": false,
-  "do_predict": false,
-  "resume_from_checkpoint": null,
-  "warmup_ratio": 0.05,
-  "logging_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_stock_32f_10pct/v0-20260606-044138/runs",
-  "local_rank": 0,
-  "sortish_sampler": false,
-  "predict_with_generate": false,
-  "generation_max_length": null,
-  "generation_num_beams": null,
-  "generation_config": null,
-  "tuner_backend": "peft",
-  "vit_gradient_checkpointing": true,
-  "router_aux_loss_coef": 0.0,
-  "enable_dft_loss": false,
-  "enable_channel_loss": false,
-  "safe_serialization": true,
-  "max_shard_size": "5GB",
-  "check_model": true,
-  "acc_strategy": "token",
-  "train_dataloader_shuffle": true,
-  "group_by_length": false,
-  "max_epochs": null,
-  "aligner_lr": null,
-  "vit_lr": 1e-06,
-  "use_logits_to_keep": null,
-  "ds3_gather_for_generation": true,
-  "resume_only_model": false,
-  "optimizer": null,
-  "loss_type": null,
-  "eval_metric": null,
-  "callbacks": [],
-  "early_stop_interval": null,
-  "eval_use_evalscope": false,
-  "eval_dataset": [],
-  "eval_dataset_args": null,
-  "eval_limit": null,
-  "eval_generation_config": null,
-  "extra_eval_args": null,
-  "tuner_type": "full",
-  "use_galore": false,
-  "galore_target_modules": null,
-  "galore_rank": 128,
-  "galore_update_proj_gap": 50,
-  "galore_scale": 1.0,
-  "galore_proj_type": "std",
-  "galore_optim_per_parameter": false,
-  "galore_with_embedding": false,
-  "galore_quantization": false,
-  "galore_proj_quant": false,
-  "galore_proj_bits": 4,
-  "galore_proj_group_size": 256,
-  "galore_cos_threshold": 0.4,
-  "galore_gamma_proj": 2,
-  "galore_queue_size": 5,
-  "lisa_activated_layers": 0,
-  "lisa_step_interval": 20,
-  "use_flash_ckpt": false,
-  "use_ray": false,
-  "ray_exp_name": null,
-  "device_groups": null,
-  "model": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_stock_a800/v0-20260529-170602/checkpoint-2181",
-  "model_type": "llava_qwen3_5vit_qwen3",
-  "model_revision": null,
-  "task_type": "causal_lm",
-  "torch_dtype": "bfloat16",
-  "attn_impl": "flash_attn",
-  "experts_impl": null,
-  "new_special_tokens": [],
-  "num_labels": null,
-  "problem_type": null,
-  "rope_scaling": null,
-  "device_map": null,
-  "max_memory": {},
-  "max_model_len": null,
-  "local_repo_path": null,
-  "init_strategy": null,
-  "template": "llava_qwen3_5vit_qwen3",
-  "system": null,
-  "max_length": 16384,
-  "truncation_strategy": "delete",
-  "max_pixels": null,
-  "agent_template": null,
-  "norm_bbox": null,
-  "use_chat_template": true,
-  "padding_side": "right",
-  "padding_free": true,
-  "loss_scale": "default",
-  "sequence_parallel_size": 1,
-  "template_backend": "swift",
-  "response_prefix": null,
-  "enable_thinking": null,
-  "add_non_thinking_prefix": true,
-  "dataset": [],
-  "val_dataset": [],
-  "cached_dataset": [
-    "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/image_10pct/train",
-    "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/video_10pct_32f/train"
-  ],
-  "cached_val_dataset": [],
-  "split_dataset_ratio": 0.0,
-  "dataset_num_proc": 16,
-  "load_from_cache_file": false,
-  "dataset_shuffle": true,
-  "val_dataset_shuffle": false,
-  "streaming": false,
-  "interleave_prob": null,
-  "stopping_strategy": "first_exhausted",
-  "shuffle_buffer_size": 1000,
-  "download_mode": "reuse_dataset_if_exists",
-  "columns": {},
-  "strict": false,
-  "model_name": null,
-  "model_author": null,
-  "custom_dataset_info": [],
-  "quant_method": null,
-  "quant_bits": null,
-  "hqq_axis": null,
-  "bnb_4bit_compute_dtype": "bfloat16",
-  "bnb_4bit_quant_type": "nf4",
-  "bnb_4bit_use_double_quant": true,
-  "bnb_4bit_quant_storage": null,
-  "max_new_tokens": 64,
-  "temperature": 0.0,
-  "top_k": null,
-  "top_p": null,
-  "repetition_penalty": null,
-  "num_beams": 1,
-  "stream": false,
-  "stop_words": [],
-  "logprobs": false,
-  "top_logprobs": null,
-  "structured_outputs_regex": null,
-  "adapters": [],
-  "external_plugins": [
-    "ms-swift/video_mllm/model_plugin_qwen3_5vit.py",
-    "ms-swift/video_mllm/dataset_plugin.py"
-  ],
-  "custom_register_path": [],
-  "model_kwargs": {},
-  "load_args": false,
-  "load_data_args": false,
-  "packing": true,
-  "packing_length": 16384,
-  "packing_num_proc": 1,
-  "lazy_tokenize": false,
-  "use_hf": true,
-  "ignore_args_error": false,
-  "use_swift_lora": false,
-  "freeze_parameters": [],
-  "freeze_parameters_regex": null,
-  "freeze_parameters_ratio": 0.0,
-  "trainable_parameters": [
-    "model.multi_modal_projector"
-  ],
-  "trainable_parameters_regex": null,
-  "freeze_llm": false,
-  "freeze_vit": false,
-  "freeze_aligner": false,
-  "target_modules": [
-    "all-linear"
-  ],
-  "target_regex": null,
-  "target_parameters": null,
-  "modules_to_save": [],
-  "lora_rank": 8,
-  "lora_alpha": 32,
-  "lora_dropout": 0.05,
-  "lora_bias": "none",
-  "lora_dtype": null,
-  "lorap_lr_ratio": null,
-  "use_rslora": false,
-  "use_dora": false,
-  "lora_ga_batch_size": 2,
-  "lora_ga_iters": 2,
-  "lora_ga_max_length": 1024,
-  "lora_ga_direction": "ArB2r",
-  "lora_ga_scale": "stable",
-  "lora_ga_stable_gamma": 16,
-  "init_weights": true,
-  "fourier_n_frequency": 2000,
-  "fourier_scaling": 300.0,
-  "boft_block_size": 4,
-  "boft_block_num": 0,
-  "boft_n_butterfly_factor": 1,
-  "boft_dropout": 0.0,
-  "vera_rank": 256,
-  "vera_projection_prng_key": 0,
-  "vera_dropout": 0.0,
-  "vera_d_initial": 0.1,
-  "adapter_act": "gelu",
-  "adapter_length": 128,
-  "adalora_target_r": 8,
-  "adalora_init_r": 12,
-  "adalora_tinit": 0,
-  "adalora_tfinal": 0,
-  "adalora_deltaT": 1,
-  "adalora_beta1": 0.85,
-  "adalora_beta2": 0.85,
-  "adalora_orth_reg_weight": 0.5,
-  "llamapro_num_new_blocks": 4,
-  "llamapro_num_groups": null,
-  "reft_layer_key": null,
-  "reft_layers": null,
-  "reft_rank": 4,
-  "reft_intervention_type": "LoreftIntervention",
-  "reft_args": null,
-  "swanlab_token": null,
-  "swanlab_project": "ms-swift",
-  "swanlab_workspace": null,
-  "swanlab_exp_name": null,
-  "swanlab_notification_method": null,
-  "swanlab_webhook_url": null,
-  "swanlab_secret": null,
-  "swanlab_sender_email": null,
-  "swanlab_receiver_email": null,
-  "swanlab_smtp_server": null,
-  "swanlab_smtp_port": null,
-  "swanlab_email_language": "zh",
-  "swanlab_mode": "cloud",
-  "add_version": true,
-  "create_checkpoint_symlink": false,
-  "zero_hpz_partition_size": null,
-  "deepspeed_autotp_size": null,
-  "swift_version": "4.1.1",
-  "ckpt_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_stock_a800/v0-20260529-170602/checkpoint-2181",
-  "rank": 0,
-  "global_world_size": 8,
-  "local_world_size": 8,
-  "model_suffix": "checkpoint-2181",
-  "model_info": "ModelInfo(model_type='llava_qwen3_5vit_qwen3', model_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_stock_a800/v0-20260529-170602/checkpoint-2181', torch_dtype=torch.bfloat16, max_model_len=262144, quant_method=None, quant_bits=None, rope_scaling=None, is_moe_model=False, is_multimodal=True, config=None, task_type='causal_lm', num_labels=None)",
-  "model_meta": "ModelMeta(model_type='llava_qwen3_5vit_qwen3', model_groups=[ModelGroup(models=[Model(ms_model_id='Qwen/Qwen3-0.6B', hf_model_id='Qwen/Qwen3-0.6B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-1.7B', hf_model_id='Qwen/Qwen3-1.7B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-4B', hf_model_id='Qwen/Qwen3-4B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-8B', hf_model_id='Qwen/Qwen3-8B', model_path=None, ms_revision=None, hf_revision=None)], template=None, ignore_patterns=None, requires=None, tags=[])], loader=<class 'model_plugin_qwen3_5vit.Qwen3_5ViTQwen3Loader'>, template='llava_qwen3_5vit_qwen3', model_arch=MultiModelKeys(arch_name='llava_hf', embedding=None, module_list=None, lm_head=None, q_proj=None, k_proj=None, v_proj=None, o_proj=None, attention=None, mlp=None, down_proj=None, qkv_proj=None, qk_proj=None, qa_proj=None, qb_proj=None, kv_proj=None, kva_proj=None, kvb_proj=None, language_model=['model.language_model', 'lm_head'], aligner=['model.multi_modal_projector'], vision_tower=['model.vision_tower'], generator=[]), mcore_model_type=None, architectures=['LlavaQwen3_5ViTForConditionalGeneration'], additional_saved_files=[], torch_dtype=None, is_multimodal=True, is_reward=False, task_type=None, ignore_patterns=None, requires=[], tags=['vision', 'video'])",
-  "model_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_stock_a800/v0-20260529-170602/checkpoint-2181",
-  "template_meta": "QwenTemplateMeta(template_type='llava_qwen3_5vit_qwen3', prefix=[], prompt=['<|im_start|>user\\n{{QUERY}}<|im_end|>\\n<|im_start|>assistant\\n'], chat_sep=['<|im_end|>\\n'], suffix=['<|im_end|>\\n'], template_cls=<class 'model_plugin_qwen3_5vit.Qwen3_5ViTLlavaTemplate'>, system_prefix=['<|im_start|>system\\n{{SYSTEM}}<|im_end|>\\n'], default_system=None, auto_add_bos=False, stop_words=['<|endoftext|>'], agent_template='hermes', is_thinking=False, thinking_prefix='', non_thinking_prefix='', history_thinking_prefix='')",
-  "_val_dataset_exists": false,
-  "hub": "<class 'swift.hub.hub.HFHub'>",
-  "evaluation_strategy": "steps",
-  "training_args": "Seq2SeqTrainingArguments(output_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_stock_32f_10pct/v0-20260606-044138', per_device_train_batch_size=1, num_train_epochs=1.0, max_steps=-1, learning_rate=1e-05, lr_scheduler_type=<SchedulerType.COSINE: 'cosine'>, lr_scheduler_kwargs=None, warmup_steps=0.05, optim=<OptimizerNames.ADAMW_TORCH_FUSED: 'adamw_torch_fused'>, optim_args=None, weight_decay=0.1, adam_beta1=0.9, adam_beta2=0.95, adam_epsilon=1e-08, optim_target_modules=None, gradient_accumulation_steps=8, average_tokens_across_devices=None, max_grad_norm=1.0, label_smoothing_factor=0.0, bf16=True, fp16=False, bf16_full_eval=False, fp16_full_eval=False, tf32=None, gradient_checkpointing=True, gradient_checkpointing_kwargs={'use_reentrant': False}, torch_compile=False, torch_compile_backend=None, torch_compile_mode=None, use_liger_kernel=False, liger_kernel_config=None, use_cache=False, neftune_noise_alpha=None, torch_empty_cache_steps=None, auto_find_batch_size=False, logging_strategy=<IntervalStrategy.STEPS: 'steps'>, logging_steps=1, logging_first_step=True, log_on_each_node=True, logging_nan_inf_filter=True, include_num_input_tokens_seen=None, log_level='passive', log_level_replica='warning', disable_tqdm=False, report_to=[], run_name='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_stock_32f_10pct/v0-20260606-044138', project='huggingface', trackio_space_id='trackio', eval_strategy=<IntervalStrategy.NO: 'no'>, eval_steps=200.0, eval_delay=0, per_device_eval_batch_size=1, prediction_loss_only=False, eval_on_start=False, eval_do_concat_batches=True, eval_use_gather_object=False, eval_accumulation_steps=None, include_for_metrics=[], batch_eval_metrics=False, save_only_model=False, save_strategy=<SaveStrategy.STEPS: 'steps'>, save_steps=200, save_on_each_node=False, save_total_limit=1, enable_jit_checkpoint=False, push_to_hub=False, hub_token=None, hub_private_repo=None, hub_model_id=None, hub_strategy=<HubStrategy.EVERY_SAVE: 'every_save'>, hub_always_push=False, hub_revision=None, load_best_model_at_end=False, metric_for_best_model='loss', greater_is_better=False, ignore_data_skip=False, restore_callback_states_from_checkpoint=False, full_determinism=False, seed=42, data_seed=42, use_cpu=False, accelerator_config=AcceleratorConfig(split_batches=False, dispatch_batches=False, even_batches=True, use_seedable_sampler=True, non_blocking=False, gradient_accumulation_kwargs=None, use_configured_state=False), parallelism_config=None, dataloader_drop_last=False, dataloader_num_workers=6, dataloader_pin_memory=True, dataloader_persistent_workers=True, dataloader_prefetch_factor=4, remove_unused_columns=False, label_names=None, train_sampling_strategy='random', length_column_name='length', ddp_find_unused_parameters=None, ddp_bucket_cap_mb=None, ddp_broadcast_buffers=None, ddp_backend=None, ddp_timeout=7200, fsdp=[], fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}, deepspeed={'fp16': {'enabled': 'auto', 'loss_scale': 0, 'loss_scale_window': 1000, 'initial_scale_power': 16, 'hysteresis': 2, 'min_loss_scale': 1}, 'bf16': {'enabled': 'auto'}, 'zero_optimization': {'stage': 2, 'offload_optimizer': {'device': 'none', 'pin_memory': True}, 'allgather_partitions': True, 'allgather_bucket_size': 200000000.0, 'overlap_comm': False, 'reduce_scatter': True, 'reduce_bucket_size': 200000000.0, 'contiguous_gradients': True}, 'gradient_accumulation_steps': 'auto', 'gradient_clipping': 'auto', 'steps_per_print': 2000, 'train_batch_size': 'auto', 'train_micro_batch_size_per_gpu': 'auto', 'wall_clock_breakdown': False}, debug=[], skip_memory_metrics=True, do_train=False, do_eval=False, do_predict=False, resume_from_checkpoint=None, warmup_ratio=0.05, logging_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_stock_32f_10pct/v0-20260606-044138/runs', local_rank=0, sortish_sampler=False, predict_with_generate=False, generation_max_length=None, generation_num_beams=None, generation_config=None, tuner_backend='peft', vit_gradient_checkpointing=True, router_aux_loss_coef=0.0, enable_dft_loss=False, enable_channel_loss=False, safe_serialization=True, max_shard_size='5GB', check_model=True, acc_strategy='token', train_dataloader_shuffle=True, group_by_length=False, max_epochs=None, aligner_lr=None, vit_lr=1e-06, use_logits_to_keep=None, ds3_gather_for_generation=True, resume_only_model=False, optimizer='multimodal', loss_type=None, eval_metric=None, callbacks=[], early_stop_interval=None, eval_use_evalscope=False, eval_dataset=[], eval_dataset_args=None, eval_limit=None, eval_generation_config=None, extra_eval_args=None, tuner_type='full', use_galore=False, galore_target_modules=None, galore_rank=128, galore_update_proj_gap=50, galore_scale=1.0, galore_proj_type='std', galore_optim_per_parameter=False, galore_with_embedding=False, galore_quantization=False, galore_proj_quant=False, galore_proj_bits=4, galore_proj_group_size=256, galore_cos_threshold=0.4, galore_gamma_proj=2, galore_queue_size=5, lisa_activated_layers=0, lisa_step_interval=20, use_flash_ckpt=False)"
-}
\ No newline at end of file
diff --git a/ckpts/stock_32f/chat_template.jinja b/ckpts/stock_32f/chat_template.jinja
deleted file mode 100644
index 70adff8a08fb31e0636f618564838d4bf3c05286..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/chat_template.jinja
+++ /dev/null
@@ -1,61 +0,0 @@
-{%- if tools %}
-    {{- '<|im_start|>system\n' }}
-    {%- if messages[0].role == 'system' %}
-        {{- messages[0].content + '\n\n' }}
-    {%- endif %}
-    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
-    {%- for tool in tools %}
-        {{- "\n" }}
-        {{- tool | tojson }}
-    {%- endfor %}
-    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
-{%- else %}
-    {%- if messages[0].role == 'system' %}
-        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
-    {%- endif %}
-{%- endif %}
-{%- for message in messages %}
-    {%- if message.content is string %}
-        {%- set content = message.content %}
-    {%- else %}
-        {%- set content = '' %}
-    {%- endif %}
-    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
-        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
-    {%- elif message.role == "assistant" %}
-        {{- '<|im_start|>' + message.role + '\n' + content }}
-        {%- if message.tool_calls %}
-            {%- for tool_call in message.tool_calls %}
-                {%- if (loop.first and content) or (not loop.first) %}
-                    {{- '\n' }}
-                {%- endif %}
-                {%- if tool_call.function %}
-                    {%- set tool_call = tool_call.function %}
-                {%- endif %}
-                {{- '<tool_call>\n{"name": "' }}
-                {{- tool_call.name }}
-                {{- '", "arguments": ' }}
-                {%- if tool_call.arguments is string %}
-                    {{- tool_call.arguments }}
-                {%- else %}
-                    {{- tool_call.arguments | tojson }}
-                {%- endif %}
-                {{- '}\n</tool_call>' }}
-            {%- endfor %}
-        {%- endif %}
-        {{- '<|im_end|>\n' }}
-    {%- elif message.role == "tool" %}
-        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
-            {{- '<|im_start|>user' }}
-        {%- endif %}
-        {{- '\n<tool_response>\n' }}
-        {{- content }}
-        {{- '\n</tool_response>' }}
-        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
-            {{- '<|im_end|>\n' }}
-        {%- endif %}
-    {%- endif %}
-{%- endfor %}
-{%- if add_generation_prompt %}
-    {{- '<|im_start|>assistant\n' }}
-{%- endif %}
\ No newline at end of file
diff --git a/ckpts/stock_32f/chat_template.json b/ckpts/stock_32f/chat_template.json
deleted file mode 100644
index 68724c68b1612e9ab2e31f84e3279099ad656ff6..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/chat_template.json
+++ /dev/null
@@ -1,3 +0,0 @@
-{
-  "chat_template": "{% for message in messages %}{{'<|im_start|>' + message['role'] + ' '}}{# Render all images first #}{% for content in message['content'] | selectattr('type', 'equalto', 'image') %}{{ '<image>' }}{% endfor %}{# Render all video then #}{% for content in message['content'] | selectattr('type', 'equalto', 'video') %}{{ '<video>' }}{% endfor %}{# Render all text next #}{% if message['role'] != 'assistant' %}{% for content in message['content'] | selectattr('type', 'equalto', 'text') %}{{ '\n' + content['text'] }}{% endfor %}{% else %}{% for content in message['content'] | selectattr('type', 'equalto', 'text') %}{% generation %}{{ '\n' + content['text'] }}{% endgeneration %}{% endfor %}{% endif %}{{'<|im_end|>'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"
-}
diff --git a/ckpts/stock_32f/config.json b/ckpts/stock_32f/config.json
deleted file mode 100644
index ce830f6986f152f6aeb43595f9a789a81c62b646..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/config.json
+++ /dev/null
@@ -1,262 +0,0 @@
-{
-  "architectures": [
-    "LlavaQwen3_5ViTForConditionalGeneration"
-  ],
-  "bos_token_id": null,
-  "dtype": "bfloat16",
-  "eos_token_id": 151645,
-  "hidden_size": 2560,
-  "image_grid_pinpoints": [
-    [
-      384,
-      384
-    ],
-    [
-      384,
-      768
-    ],
-    [
-      384,
-      1152
-    ],
-    [
-      384,
-      1536
-    ],
-    [
-      384,
-      1920
-    ],
-    [
-      384,
-      2304
-    ],
-    [
-      768,
-      384
-    ],
-    [
-      768,
-      768
-    ],
-    [
-      768,
-      1152
-    ],
-    [
-      768,
-      1536
-    ],
-    [
-      768,
-      1920
-    ],
-    [
-      768,
-      2304
-    ],
-    [
-      1152,
-      384
-    ],
-    [
-      1152,
-      768
-    ],
-    [
-      1152,
-      1152
-    ],
-    [
-      1152,
-      1536
-    ],
-    [
-      1152,
-      1920
-    ],
-    [
-      1152,
-      2304
-    ],
-    [
-      1536,
-      384
-    ],
-    [
-      1536,
-      768
-    ],
-    [
-      1536,
-      1152
-    ],
-    [
-      1536,
-      1536
-    ],
-    [
-      1536,
-      1920
-    ],
-    [
-      1536,
-      2304
-    ],
-    [
-      1920,
-      384
-    ],
-    [
-      1920,
-      768
-    ],
-    [
-      1920,
-      1152
-    ],
-    [
-      1920,
-      1536
-    ],
-    [
-      1920,
-      1920
-    ],
-    [
-      1920,
-      2304
-    ],
-    [
-      2304,
-      384
-    ],
-    [
-      2304,
-      768
-    ],
-    [
-      2304,
-      1152
-    ],
-    [
-      2304,
-      1536
-    ],
-    [
-      2304,
-      1920
-    ],
-    [
-      2304,
-      2304
-    ]
-  ],
-  "image_token_index": 151669,
-  "model_type": "llava_qwen3_5vit_qwen3",
-  "multimodal_projector_bias": true,
-  "pad_token_id": 151643,
-  "projector_hidden_act": "gelu",
-  "text_config": {
-    "_name_or_path": "/share/m2v_intern_v3/wangjunjie09/model_cache/huggingface/Qwen/Qwen3-4B-Instruct-2507",
-    "architectures": [
-      "Qwen3ForCausalLM"
-    ],
-    "attention_bias": false,
-    "attention_dropout": 0.0,
-    "bos_token_id": 151643,
-    "dtype": "bfloat16",
-    "eos_token_id": 151645,
-    "head_dim": 128,
-    "hidden_act": "silu",
-    "hidden_size": 2560,
-    "initializer_range": 0.02,
-    "intermediate_size": 9728,
-    "layer_types": [
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention"
-    ],
-    "max_position_embeddings": 262144,
-    "max_window_layers": 36,
-    "model_type": "qwen3",
-    "num_attention_heads": 32,
-    "num_hidden_layers": 36,
-    "num_key_value_heads": 8,
-    "pad_token_id": 151643,
-    "rms_norm_eps": 1e-06,
-    "rope_parameters": {
-      "rope_theta": 5000000,
-      "rope_type": "default"
-    },
-    "sliding_window": null,
-    "tie_word_embeddings": true,
-    "use_cache": false,
-    "use_sliding_window": false,
-    "vocab_size": 151936
-  },
-  "tie_word_embeddings": true,
-  "transformers_version": "5.5.4",
-  "use_cache": false,
-  "video_token_index": 151670,
-  "vision_aspect_ratio": "anyres_max_9",
-  "vision_config": {
-    "deepstack_visual_indexes": [],
-    "depth": 24,
-    "dtype": "bfloat16",
-    "hidden_act": "gelu_pytorch_tanh",
-    "hidden_size": 1024,
-    "image_size": 384,
-    "in_channels": 3,
-    "initializer_range": 0.02,
-    "intermediate_size": 4096,
-    "model_type": "qwen3_5",
-    "num_heads": 16,
-    "num_position_embeddings": 2304,
-    "out_hidden_size": 2048,
-    "patch_size": 16,
-    "spatial_merge_size": 2,
-    "temporal_patch_size": 2
-  },
-  "vision_feature_layer": -1,
-  "vision_feature_select_strategy": "full",
-  "auto_map": {
-    "AutoConfig": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTConfig",
-    "AutoModel": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
-    "AutoModelForCausalLM": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
-    "AutoModelForImageTextToText": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration"
-  }
-}
\ No newline at end of file
diff --git a/ckpts/stock_32f/generation_config.json b/ckpts/stock_32f/generation_config.json
deleted file mode 100644
index 56320d3179df3512d0cb76c240831a10e6f7c6f7..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/generation_config.json
+++ /dev/null
@@ -1,12 +0,0 @@
-{
-  "_from_model_config": true,
-  "bos_token_id": 151643,
-  "eos_token_id": [
-    151645,
-    151643
-  ],
-  "output_attentions": false,
-  "output_hidden_states": false,
-  "transformers_version": "5.5.4",
-  "use_cache": true
-}
diff --git a/ckpts/stock_32f/latest b/ckpts/stock_32f/latest
deleted file mode 100644
index 2c429357c7f2fba84efb27b0672c454288951b67..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/latest
+++ /dev/null
@@ -1 +0,0 @@
-global_step785
\ No newline at end of file
diff --git a/ckpts/stock_32f/model-00001-of-00002.safetensors b/ckpts/stock_32f/model-00001-of-00002.safetensors
deleted file mode 100644
index 64a14f499bf8ca4ca6bfe8cae9061f5e06d3b26c..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/model-00001-of-00002.safetensors
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:8bb89a0cb17b390ed88d5b6c0b432dd14ec2f7e68fcfe1baf2db3ed549fcaf25
-size 4987507408
diff --git a/ckpts/stock_32f/model-00002-of-00002.safetensors b/ckpts/stock_32f/model-00002-of-00002.safetensors
deleted file mode 100644
index 18ab40420122e710ec8b12d349b184f671ec1c34..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/model-00002-of-00002.safetensors
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:549730da56dab26ad27866c5c2167672c43a33e049aac3c1e96eb2cd765cb529
-size 4466285248
diff --git a/ckpts/stock_32f/model.safetensors.index.json b/ckpts/stock_32f/model.safetensors.index.json
deleted file mode 100644
index a227833b5ba2c2ae9deeddde8d4fc716babffc8c..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/model.safetensors.index.json
+++ /dev/null
@@ -1,705 +0,0 @@
-{
-  "metadata": {
-    "total_parameters": 4337895424,
-    "total_size": 9453703168
-  },
-  "weight_map": {
-    "lm_head.weight": "model-00001-of-00002.safetensors",
-    "model.image_newline": "model-00001-of-00002.safetensors",
-    "model.language_model.embed_tokens.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.norm.weight": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_1.bias": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_1.weight": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_2.bias": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.final_layernorm.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.final_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.patch_embed.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.patch_embed.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.pos_embed.weight": "model-00002-of-00002.safetensors"
-  }
-}
diff --git a/ckpts/stock_32f/modeling_qwen3_5vit_qwen3.py b/ckpts/stock_32f/modeling_qwen3_5vit_qwen3.py
deleted file mode 100644
index 05d150ac147b78e77aff902bda1daf0c26759541..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/modeling_qwen3_5vit_qwen3.py
+++ /dev/null
@@ -1,405 +0,0 @@
-"""
-modeling_qwen3_5vit_qwen3.py — Qwen3.5 Vision as SigLIP-compat vision_tower in LlavaOnevision.
-
-设计：与 `modeling_qwen3vlvit_qwen3.py` 严格同构，仅换 vision backbone 源：
-  - Qwen3VLVisionModel → Qwen3_5VisionModel（继承关系：Qwen3_5VisionModel(Qwen3VLVisionModel) 去 DeepStack）
-  - Qwen3VLVisionConfig → Qwen3_5VisionConfig（父类用 AttributeError 哨兵屏蔽 deepstack_visual_indexes）
-
-其余（Adapter 契约翻译、MLP projector + pre_norm、LlavaOnevision 继承 wire class）与
-Qwen3-VL ViT pipeline 完全一致。两条 pipeline 并存意义：DeepStack ablation 天然实验组。
-
-类层级：
-  Qwen3_5ViTBackbone(Qwen3_5VisionModel)      — 去 merger，保持 NaViT 契约
-  Qwen3_5ViTAsSiglipAdapter(nn.Module)        — 持有 Backbone，做 SigLIP ↔ NaViT 契约翻译
-
-三方对比公平性：定 384×384 AnyRes tile + 同款 projector 骨架 + 同款 Qwen3-1.7B LLM。
-"""
-
-import math
-import os
-import sys
-from typing import Optional
-
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-from transformers import (
-    AutoConfig,
-    AutoModel,
-    AutoModelForCausalLM,
-    LlavaOnevisionConfig,
-    LlavaOnevisionForConditionalGeneration,
-    LlavaOnevisionModel,
-    LlavaOnevisionPreTrainedModel,
-    Qwen3Config,
-)
-from transformers.activations import ACT2FN
-from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
-from transformers.models.qwen3_5.configuration_qwen3_5 import Qwen3_5VisionConfig
-from transformers.models.qwen3_5.modeling_qwen3_5 import Qwen3_5VisionModel
-
-# Shared layout-permutation utility lives in declip_qwenvit (single source of
-# truth — same code path runs in declip-training-side qk_cosine reorder).
-# Add VisionEncoder repo root to sys.path so this modeling file is importable
-# even when the package isn't pip-installed (ms-swift integration loads it
-# via dynamic plugin path).
-_REPO_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
-if _REPO_ROOT not in sys.path:
-    sys.path.append(_REPO_ROOT)
-from declip_qwenvit.model.qwen3vit_qk import block_merge_to_row_major_permutation  # noqa: E402
-
-
-class LlavaQwen3_5ViTConfig(LlavaOnevisionConfig):
-    """LlavaOnevisionConfig 子类，vision_config 类型换成 Qwen3_5VisionConfig。
-
-    与 Qwen3-VL ViT 版的差异：
-      - sub_configs.vision_config 用 Qwen3_5VisionConfig
-      - 不再设置 deepstack_visual_indexes（Qwen3_5VisionConfig 用 AttributeError 哨兵屏蔽此字段）
-
-    其余同 LlavaQwen3VLViTConfig（tile_size 默认 384；vision_feature_select_strategy='full'
-    必须 override，Qwen3.5 ViT 无 CLS token）。
-    """
-
-    model_type = "llava_qwen3_5vit_qwen3"
-    sub_configs = {"vision_config": Qwen3_5VisionConfig, "text_config": Qwen3Config}
-
-    def __init__(
-        self,
-        vision_config=None,
-        text_config=None,
-        tile_size: int = 384,
-        vit_register_ring: int = 0,
-        vit_skip_last_blocks: int = 0,
-        **kwargs,
-    ):
-        if isinstance(vision_config, dict):
-            vision_config = Qwen3_5VisionConfig(**vision_config)
-        elif vision_config is None:
-            vision_config = Qwen3_5VisionConfig()
-        # UniRefiner deploy-side register ring width in tokens (V10.0.1+). The S0
-        # refinement optimizes image tokens INSIDE a register-ring canvas; >0 makes
-        # every tower forward pad a zero ring around each tile/frame and drop the
-        # ring tokens after encoding (LLM-side token contract unchanged). Persisted
-        # in config.json so S2-resume and eval inherit it automatically. 0 = stock
-        # behavior (all non-V10.0.1 chains).
-        self.vit_register_ring = int(vit_register_ring)
-        # -N layer ablation (stock_minus2 chains, 2026-06-10): physically DELETE the
-        # last N ViT blocks after construction (LLaVA-NeXT convention,
-        # `del encoder.layers[-1:]`), so last_hidden_state = hidden_states[-(N+1)]
-        # of the full tower, still through final_layernorm. Physical deletion (vs
-        # forward-skip) keeps the state_dict 1:1 with truncated declip S0 ckpts
-        # (missing=0 contract) and makes silently running the dropped block
-        # impossible. Persisted in config.json; S2-resume/eval inherit. 0 = stock.
-        # NOTE: vision_config.depth stays at the full count; truncation applies at
-        # adapter build, before any load_state_dict.
-        self.vit_skip_last_blocks = int(vit_skip_last_blocks)
-        # WHY 无 `vision_config.deepstack_visual_indexes = []`（对比 Qwen3-VL ViT 版）：
-        # Qwen3_5VisionConfig 父类用 AttributeError() 哨兵显式屏蔽此字段，设置会报 AttributeError
-        # LlavaOnevision.pack_image_features 用这个作为 tile 像素大小（不是 patch_size）
-        vision_config.image_size = tile_size
-
-        if isinstance(text_config, dict):
-            text_config = Qwen3Config(**text_config)
-        elif text_config is None:
-            text_config = Qwen3Config()
-
-        # 父类默认 select_strategy='default' 会跳首 token (CLS) — Qwen3.5 ViT 无 CLS 必须用 'full'
-        kwargs.setdefault("vision_feature_select_strategy", "full")
-        super().__init__(vision_config=vision_config, text_config=text_config, **kwargs)
-
-
-class Qwen3_5ViTBackbone(Qwen3_5VisionModel):
-    """Qwen3.5 Vision 去除原生 patch_merger 的 backbone 版本（V6 final_layernorm fix (2026-05-16): append final_layernorm）。
-
-    构造时把 merger.norm 的预训练权重抠到 final_layernorm，然后 `del self.merger`
-    释放 ~37M 参数（保留 norm 的 LN 焊到末端做 post_layernorm 角色，丢弃 spatial
-    shuffle + linear_fc1/fc2，那对应 LlavaOV projector 的职责）。
-
-    架构对称（V6 final_layernorm fix (2026-05-16) 修复）：
-      SigLIP2:    encoder → post_layernorm → last_hidden_state → LlavaOV MLP → LLM
-      V6 final_layernorm fix (2026-05-16):     encoder → final_layernorm → last_hidden_state → LlavaOV MLP → LLM
-
-    forward 跑完 transformer blocks 后过 final_layernorm，再返回。下游 LlavaOnevision
-    pack_image_features 的 AnyRes 2×2 pool 接管原 merger 的空间合并职责。
-
-    输入输出契约与父类 Qwen3_5VisionModel 一致（NaViT flat）：
-      forward(hidden_states=[L, patch_dim], grid_thw=[N, 3])
-      → BaseModelOutput(last_hidden_state=[L, hidden_size])
-
-    forward 主体 1:1 对照 `Qwen3_5VisionModel.forward`（已无 deepstack loop，比
-    Qwen3VLVisionModel.forward 更短），仅跳过末尾 `self.merger(x)`，改为 final_layernorm。
-    """
-
-    def __init__(self, config):
-        super().__init__(config)
-        # V6 final_layernorm fix (2026-05-16): extract merger.norm pretrained weights into final_layernorm.
-        # Default init: even if ckpt 阶段没 inject final_layernorm.* (e.g. stock
-        # bootstrap path that filters merger.*), final_layernorm 仍持有 merger.norm
-        # 的预训练值, 不是 random — 这是防止 silent corruption 的兜底.
-        ln_w = self.merger.norm.weight.detach().clone()
-        ln_b = self.merger.norm.bias.detach().clone()
-        del self.merger
-        self.final_layernorm = nn.LayerNorm(config.hidden_size, eps=1e-6)
-        self.final_layernorm.weight.data.copy_(ln_w)
-        self.final_layernorm.bias.data.copy_(ln_b)
-
-    def forward(self, hidden_states, grid_thw, **kwargs):
-        hidden_states = self.patch_embed(hidden_states)
-
-        pos_embeds = self.fast_pos_embed_interpolate(grid_thw)
-        hidden_states = hidden_states + pos_embeds
-
-        rotary_pos_emb = self.rot_pos_emb(grid_thw)
-        seq_len, _ = hidden_states.size()
-        rotary_pos_emb = rotary_pos_emb.reshape(seq_len, -1)
-        emb = torch.cat((rotary_pos_emb, rotary_pos_emb), dim=-1)
-        position_embeddings = (emb.cos(), emb.sin())
-
-        cu_seqlens = torch.repeat_interleave(
-            grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]
-        ).cumsum(dim=0, dtype=torch.int32)
-        cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
-
-        for blk in self.blocks:
-            hidden_states = blk(
-                hidden_states,
-                cu_seqlens=cu_seqlens,
-                position_embeddings=position_embeddings,
-                **kwargs,
-            )
-
-        # V6 final_layernorm fix (2026-05-16): appended final LayerNorm — mirrors SigLIP2's post_layernorm.
-        # Per-token affine; layout-invariant (reorder happens in adapter).
-        hidden_states = self.final_layernorm(hidden_states)
-
-        return BaseModelOutput(last_hidden_state=hidden_states)
-
-
-class Qwen3_5ViTAsSiglipAdapter(nn.Module):
-    """SigLIP 契约 → NaViT 契约的翻译层。持有 Qwen3_5ViTBackbone。
-
-    对外暴露 SigLIP 式 forward(pixel_values=[N,3,H,W]) → BaseModelOutputWithPooling，
-    供 LlavaOnevision 消费；对内按官方 _preprocess 的 reshape 链把 pixel_values
-    转成 NaViT flat + grid_thw 喂给 Backbone。
-
-    reshape 链 1:1 照抄 transformers 官方 Qwen2VLImageProcessorFast._preprocess
-    (video_processing_qwen3_vl.py L227-252) —— Qwen3.5 无独立 image_processor，复用 Qwen3-VL 格式。
-    """
-
-    def __init__(self, vision_config: Qwen3_5VisionConfig, register_ring: int = 0,
-                 skip_last_blocks: int = 0):
-        super().__init__()
-        self.vision = Qwen3_5ViTBackbone(vision_config)
-        self.config = vision_config
-        # Deploy-side UniRefiner register ring (tokens). See LlavaQwen3_5ViTConfig.
-        # Ring must keep the padded grid even for spatial_merge_size=2 NaViT
-        # flattening: 24 + 2*ring stays even for any integer ring.
-        self.register_ring = int(register_ring)
-        # -N layer ablation: physically delete the last N blocks (see
-        # LlavaQwen3_5ViTConfig.vit_skip_last_blocks). final_layernorm stays,
-        # applied to the new last block's output.
-        skip = int(skip_last_blocks)
-        if skip > 0:
-            self.vision.blocks = self.vision.blocks[:-skip]
-        self.skip_last_blocks = skip
-
-    @property
-    def dtype(self):
-        return next(self.parameters()).dtype
-
-    @property
-    def device(self):
-        return next(self.parameters()).device
-
-    def _flatten_navit(self, pixel_values: torch.Tensor):
-        """[N, 3, H, W] → (flat=[N*L, patch_dim], grid_thw=[N, 3], shape=(N, L)).
-
-        L = grid_t * grid_h * grid_w = 1 * (H/16) * (W/16)
-        patch_dim = C * temporal_patch_size * patch_size^2 = 3 * 2 * 16 * 16 = 1536
-        """
-        pixel_values = pixel_values.to(dtype=self.dtype)
-        tps = self.config.temporal_patch_size
-        ps = self.config.patch_size
-        ms = self.config.spatial_merge_size
-
-        patches = pixel_values.unsqueeze(1)
-        # 对单帧图像 T=1, pad=1 → expand 一帧使 T 整除 temporal_patch_size，
-        # Conv3d 在复制帧上退化为等效 2D Conv（数学无损）
-        T = patches.shape[1]
-        pad = -T % tps
-        if pad:
-            repeats = patches[:, -1:].expand(-1, pad, -1, -1, -1)
-            patches = torch.cat((patches, repeats), dim=1)
-
-        batch_size, t, channel, H, W = patches.shape
-        grid_t = t // tps
-        grid_h = H // ps
-        grid_w = W // ps
-
-        patches = patches.view(
-            batch_size, grid_t, tps, channel,
-            grid_h // ms, ms, ps,
-            grid_w // ms, ms, ps,
-        )
-        patches = patches.permute(0, 1, 4, 7, 5, 8, 3, 2, 6, 9)
-        flatten_patches = patches.reshape(
-            batch_size,
-            grid_t * grid_h * grid_w,
-            channel * tps * ps * ps,
-        )
-
-        seq_len = grid_t * grid_h * grid_w
-        flat = flatten_patches.reshape(batch_size * seq_len, -1)
-        # on-device 构造小 tensor 再 expand，host→GPU 同步量 O(3) 而非 O(N*3)
-        grid_unit = torch.tensor(
-            [grid_t, grid_h, grid_w], dtype=torch.int32, device=pixel_values.device,
-        )
-        grid_thw = grid_unit.unsqueeze(0).expand(batch_size, -1).contiguous()
-        return flat, grid_thw, (batch_size, seq_len)
-
-    def forward(
-        self,
-        pixel_values: torch.Tensor,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-        **kwargs,
-    ) -> BaseModelOutputWithPooling:
-        ring = self.register_ring
-        if ring > 0:
-            # UniRefiner deploy protocol: surround each tile/frame with a register
-            # ring, encode, drop the ring afterwards. zero-fill in normalized pixel
-            # space == the S0 training ring (register_fill: zero, gray patches).
-            pad_px = ring * self.config.patch_size
-            pixel_values = F.pad(pixel_values, (pad_px, pad_px, pad_px, pad_px), value=0.0)
-
-        flat, grid_thw, (N, S) = self._flatten_navit(pixel_values)
-        vision_out = self.vision(flat, grid_thw=grid_thw)
-        hidden = vision_out.last_hidden_state.view(N, S, -1)
-
-        # Block-merge → row-major reorder before handing to LlavaOnevision.
-        # Internally the ViT runs in Qwen NaViT block-merge layout (pretrained
-        # pos_embed + RoPE contract); downstream LlavaOV `pack_image_features`
-        # (multi-tile AnyRes path, view(num_patch_h, num_patch_w, h, w, -1))
-        # and `apply_pooling` (video path, view(B, h, w, -1) + bilinear) BOTH
-        # assume row-major. Without this reorder, the multi-tile/video spatial
-        # pool pulls together tokens that are NOT spatially adjacent — silent
-        # corruption that doesn't fire on S1 single-tile path (line 348-351 of
-        # modeling_llava_onevision.py just flattens [N,D] verbatim) but kills
-        # S2 / eval quality.
-        grid_h = int(grid_thw[0, 1].item())
-        grid_w = int(grid_thw[0, 2].item())
-        ms = getattr(self.config, "spatial_merge_size", 2)
-        layout_perm = block_merge_to_row_major_permutation(
-            grid_h, grid_w, ms=ms, device=hidden.device,
-        )
-        hidden = hidden[:, layout_perm, :]
-
-        if ring > 0:
-            # Drop the ring tokens after the row-major reorder: keep the inner
-            # (grid_h-2r)x(grid_w-2r) block. Downstream LlavaOV consumes the same
-            # row-major image-token count as the ring-off path.
-            keep = torch.ones(grid_h, grid_w, dtype=torch.bool, device=hidden.device)
-            keep[:ring, :] = False
-            keep[grid_h - ring:, :] = False
-            keep[:, :ring] = False
-            keep[:, grid_w - ring:] = False
-            hidden = hidden[:, keep.reshape(-1), :]
-
-        return BaseModelOutputWithPooling(
-            last_hidden_state=hidden,
-            # LlavaOnevision 索引 hidden_states[vision_feature_layer=-1]；tuple 长度 1 足够
-            hidden_states=(hidden,),
-            pooler_output=None,
-        )
-
-
-class LlavaQwen3_5ViTMultiModalProjector(nn.Module):
-    """标准 LlavaOnevision projector（V6 final_layernorm fix (2026-05-16): pre_norm → Identity）。
-
-    V6 final_layernorm fix (2026-05-16) 修复后, encoder 末端已自带 final_layernorm（与 SigLIP2 post_layernorm 对称），
-    projector 不再需要补 LN — pre_norm 改为 nn.Identity，对齐 SigLIP2 plugin 的
-    LlavaOnevision stock projector 结构（裸 linear_1 → GELU → linear_2），
-    保证 SigLIP2 / Qwen3.5 / Qwen3-VL 三个 backbone 在 LlavaOV 设定下公平对比。
-
-    （历史：V6.0.0~V6.0.4 时期 encoder 无 final LN，projector pre_norm 是补丁；
-    现在补丁回到 encoder 内部，projector 回归 stock 形态。）
-    """
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        super().__init__()
-        num_feature_layers = (
-            1 if isinstance(config.vision_feature_layer, int) else len(config.vision_feature_layer)
-        )
-        vision_dim = config.vision_config.hidden_size * num_feature_layers
-        text_dim = config.text_config.hidden_size
-        bias = getattr(config, "multimodal_projector_bias", True)
-
-        # V6 final_layernorm fix (2026-05-16): pre_norm = Identity (encoder 已自带 final_layernorm).
-        self.pre_norm = nn.Identity()
-        self.linear_1 = nn.Linear(vision_dim, text_dim, bias=bias)
-        self.act = ACT2FN[config.projector_hidden_act]
-        self.linear_2 = nn.Linear(text_dim, text_dim, bias=bias)
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        return self.linear_2(self.act(self.linear_1(self.pre_norm(x))))
-
-
-class LlavaQwen3_5ViTModel(LlavaOnevisionModel):
-    """继承 LlavaOnevisionModel 但绕过其 __init__ 手动装配。
-
-    父类 __init__ 调 `AutoModel.from_config(config.vision_config)` 会对 Qwen3_5VisionConfig
-    抛 "Unrecognized configuration"（Qwen3.5 vision 没注册到 AutoModel）。手动装配避开
-    这一步，同时省掉"先构造 Qwen3_5VisionModel 再被替换"的双重开销（~1.3GB init-peak）。
-
-    装配顺序与父类一致：vision_tower / projector / image_newline / language_model / post_init。
-    """
-
-    config_class = LlavaQwen3_5ViTConfig
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        # 跳过 LlavaOnevisionModel.__init__（AutoModel 不识别 Qwen3_5VisionConfig）
-        LlavaOnevisionPreTrainedModel.__init__(self, config)
-        self.vision_tower = Qwen3_5ViTAsSiglipAdapter(
-            config.vision_config,
-            register_ring=getattr(config, "vit_register_ring", 0),
-            skip_last_blocks=getattr(config, "vit_skip_last_blocks", 0),
-        )
-        self.multi_modal_projector = LlavaQwen3_5ViTMultiModalProjector(config)
-        embed_std = 1 / math.sqrt(config.text_config.hidden_size)
-        self.image_newline = nn.Parameter(
-            torch.randn(config.text_config.hidden_size, dtype=self.dtype) * embed_std
-        )
-        self.vocab_size = config.text_config.vocab_size
-        self.language_model = AutoModel.from_config(config.text_config)
-        self.post_init()
-
-
-class LlavaQwen3_5ViTForConditionalGeneration(LlavaOnevisionForConditionalGeneration):
-    """继承 LlavaOnevisionForConditionalGeneration，只换 self.model。
-
-    同样跳过父类 __init__（避免重复构造 LlavaOnevisionModel，根因见 LlavaQwen3_5ViTModel）。
-    """
-
-    config_class = LlavaQwen3_5ViTConfig
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        LlavaOnevisionPreTrainedModel.__init__(self, config)
-        self.model = LlavaQwen3_5ViTModel(config)
-        self.lm_head = nn.Linear(
-            config.text_config.hidden_size, config.text_config.vocab_size, bias=False
-        )
-        self.post_init()
-
-
-AutoConfig.register(LlavaQwen3_5ViTConfig.model_type, LlavaQwen3_5ViTConfig)
-AutoModelForCausalLM.register(LlavaQwen3_5ViTConfig, LlavaQwen3_5ViTForConditionalGeneration)
-
-
-__all__ = [
-    "LlavaQwen3_5ViTConfig",
-    "Qwen3_5ViTBackbone",
-    "Qwen3_5ViTAsSiglipAdapter",
-    "LlavaQwen3_5ViTMultiModalProjector",
-    "LlavaQwen3_5ViTModel",
-    "LlavaQwen3_5ViTForConditionalGeneration",
-]
diff --git a/ckpts/stock_32f/processor_config.json b/ckpts/stock_32f/processor_config.json
deleted file mode 100644
index 204ed5d216c648cbd71da6c25d67b4bf1ec84433..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/processor_config.json
+++ /dev/null
@@ -1,203 +0,0 @@
-{
-  "image_processor": {
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_pad": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "image_grid_pinpoints": [
-      [
-        384,
-        384
-      ],
-      [
-        384,
-        768
-      ],
-      [
-        384,
-        1152
-      ],
-      [
-        384,
-        1536
-      ],
-      [
-        384,
-        1920
-      ],
-      [
-        384,
-        2304
-      ],
-      [
-        768,
-        384
-      ],
-      [
-        768,
-        768
-      ],
-      [
-        768,
-        1152
-      ],
-      [
-        768,
-        1536
-      ],
-      [
-        768,
-        1920
-      ],
-      [
-        768,
-        2304
-      ],
-      [
-        1152,
-        384
-      ],
-      [
-        1152,
-        768
-      ],
-      [
-        1152,
-        1152
-      ],
-      [
-        1152,
-        1536
-      ],
-      [
-        1152,
-        1920
-      ],
-      [
-        1152,
-        2304
-      ],
-      [
-        1536,
-        384
-      ],
-      [
-        1536,
-        768
-      ],
-      [
-        1536,
-        1152
-      ],
-      [
-        1536,
-        1536
-      ],
-      [
-        1536,
-        1920
-      ],
-      [
-        1536,
-        2304
-      ],
-      [
-        1920,
-        384
-      ],
-      [
-        1920,
-        768
-      ],
-      [
-        1920,
-        1152
-      ],
-      [
-        1920,
-        1536
-      ],
-      [
-        1920,
-        1920
-      ],
-      [
-        1920,
-        2304
-      ],
-      [
-        2304,
-        384
-      ],
-      [
-        2304,
-        768
-      ],
-      [
-        2304,
-        1152
-      ],
-      [
-        2304,
-        1536
-      ],
-      [
-        2304,
-        1920
-      ],
-      [
-        2304,
-        2304
-      ]
-    ],
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_processor_type": "LlavaOnevisionImageProcessor",
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "size": {
-      "height": 384,
-      "width": 384
-    }
-  },
-  "image_token": "<image>",
-  "num_image_tokens": 576,
-  "processor_class": "LlavaOnevisionProcessor",
-  "video_processor": {
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "do_sample_frames": false,
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "return_metadata": false,
-    "size": {
-      "height": 384,
-      "width": 384
-    },
-    "video_processor_type": "LlavaOnevisionVideoProcessor"
-  },
-  "video_token": "<video>",
-  "vision_aspect_ratio": "anyres_max_9",
-  "vision_feature_select_strategy": null
-}
diff --git a/ckpts/stock_32f/tokenizer.json b/ckpts/stock_32f/tokenizer.json
deleted file mode 100644
index c40d89c9226d696c215d65a68eea68c6dd55d79e..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/tokenizer.json
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:9024318c850eaadf26be79389d21b07a7afd8f1af749b89f9109b06c0d12173c
-size 11423018
diff --git a/ckpts/stock_32f/tokenizer_config.json b/ckpts/stock_32f/tokenizer_config.json
deleted file mode 100644
index 91a90671a687fb5d1fc3ed576d216e5e67d9e11b..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/tokenizer_config.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "add_prefix_space": false,
-  "backend": "tokenizers",
-  "bos_token": null,
-  "clean_up_tokenization_spaces": false,
-  "eos_token": "<|im_end|>",
-  "errors": "replace",
-  "extra_special_tokens": [
-    "<image>",
-    "<video>"
-  ],
-  "is_local": true,
-  "model_max_length": 1010000,
-  "pad_token": "<|endoftext|>",
-  "processor_class": "LlavaOnevisionProcessor",
-  "split_special_tokens": false,
-  "tokenizer_class": "Qwen2Tokenizer",
-  "unk_token": null
-}
diff --git a/ckpts/stock_32f/zero_to_fp32.py b/ckpts/stock_32f/zero_to_fp32.py
deleted file mode 100644
index 5995d6e6f04e43b989587aa9022a3aef0c66d694..0000000000000000000000000000000000000000
--- a/ckpts/stock_32f/zero_to_fp32.py
+++ /dev/null
@@ -1,760 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright (c) Microsoft Corporation.
-# SPDX-License-Identifier: Apache-2.0
-
-# DeepSpeed Team
-
-# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
-# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
-# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
-# application.
-#
-# example:
-#   python zero_to_fp32.py . output_dir/
-#   or
-#   python zero_to_fp32.py . output_dir/ --safe_serialization
-
-import argparse
-import torch
-import glob
-import math
-import os
-import re
-import gc
-import json
-import numpy as np
-from tqdm import tqdm
-from collections import OrderedDict
-from dataclasses import dataclass
-
-# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
-# DeepSpeed data structures it has to be available in the current python environment.
-from deepspeed.utils import logger
-from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
-                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
-                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
-
-
-@dataclass
-class zero_model_state:
-    buffers: dict()
-    param_shapes: dict()
-    shared_params: list
-    ds_version: int
-    frozen_param_shapes: dict()
-    frozen_param_fragments: dict()
-
-
-debug = 0
-
-# load to cpu
-device = torch.device('cpu')
-
-
-def atoi(text):
-    return int(text) if text.isdigit() else text
-
-
-def natural_keys(text):
-    '''
-    alist.sort(key=natural_keys) sorts in human order
-    http://nedbatchelder.com/blog/200712/human_sorting.html
-    (See Toothy's implementation in the comments)
-    '''
-    return [atoi(c) for c in re.split(r'(\d+)', text)]
-
-
-def get_model_state_file(checkpoint_dir, zero_stage):
-    if not os.path.isdir(checkpoint_dir):
-        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
-
-    # there should be only one file
-    if zero_stage <= 2:
-        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
-    elif zero_stage == 3:
-        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
-
-    if not os.path.exists(file):
-        raise FileNotFoundError(f"can't find model states file at '{file}'")
-
-    return file
-
-
-def get_checkpoint_files(checkpoint_dir, glob_pattern):
-    # XXX: need to test that this simple glob rule works for multi-node setup too
-    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
-
-    if len(ckpt_files) == 0:
-        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
-
-    return ckpt_files
-
-
-def get_optim_files(checkpoint_dir):
-    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
-
-
-def get_model_state_files(checkpoint_dir):
-    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
-
-
-def parse_model_states(files):
-    zero_model_states = []
-    for file in files:
-        state_dict = torch.load(file, map_location=device, weights_only=False)
-
-        if BUFFER_NAMES not in state_dict:
-            raise ValueError(f"{file} is not a model state checkpoint")
-        buffer_names = state_dict[BUFFER_NAMES]
-        if debug:
-            print("Found buffers:", buffer_names)
-
-        # recover just the buffers while restoring them to fp32 if they were saved in fp16
-        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
-        param_shapes = state_dict[PARAM_SHAPES]
-
-        # collect parameters that are included in param_shapes
-        param_names = []
-        for s in param_shapes:
-            for name in s.keys():
-                param_names.append(name)
-
-        # update with frozen parameters
-        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
-        if frozen_param_shapes is not None:
-            if debug:
-                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
-            param_names += list(frozen_param_shapes.keys())
-
-        # handle shared params
-        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
-
-        ds_version = state_dict.get(DS_VERSION, None)
-
-        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
-
-        z_model_state = zero_model_state(buffers=buffers,
-                                         param_shapes=param_shapes,
-                                         shared_params=shared_params,
-                                         ds_version=ds_version,
-                                         frozen_param_shapes=frozen_param_shapes,
-                                         frozen_param_fragments=frozen_param_fragments)
-        zero_model_states.append(z_model_state)
-
-    return zero_model_states
-
-
-def parse_optim_states(files, ds_checkpoint_dir):
-    total_files = len(files)
-    state_dicts = []
-    for f in tqdm(files, desc='Loading checkpoint shards'):
-        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
-        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
-        # and also handle the case where it was already removed by another helper script
-        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
-        state_dicts.append(state_dict)
-
-    if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
-        raise ValueError(f"{files[0]} is not a zero checkpoint")
-    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
-    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
-
-    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
-    # parameters can be different from data parallelism for non-expert parameters. So we can just
-    # use the max of the partition_count to get the dp world_size.
-
-    if type(world_size) is list:
-        world_size = max(world_size)
-
-    if world_size != total_files:
-        raise ValueError(
-            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
-            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
-        )
-
-    # the groups are named differently in each stage
-    if zero_stage <= 2:
-        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
-    elif zero_stage == 3:
-        fp32_groups_key = FP32_FLAT_GROUPS
-    else:
-        raise ValueError(f"unknown zero stage {zero_stage}")
-
-    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
-    return zero_stage, world_size, fp32_flat_groups
-
-
-def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
-    """
-    Returns fp32 state_dict reconstructed from ds checkpoint
-
-    Args:
-        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
-
-    """
-    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
-
-    optim_files = get_optim_files(ds_checkpoint_dir)
-    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
-    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
-
-    model_files = get_model_state_files(ds_checkpoint_dir)
-
-    zero_model_states = parse_model_states(model_files)
-    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
-
-    if zero_stage <= 2:
-        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                                          exclude_frozen_parameters)
-    elif zero_stage == 3:
-        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                                          exclude_frozen_parameters)
-
-
-def _zero2_merge_frozen_params(state_dict, zero_model_states):
-    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
-        return
-
-    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
-    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
-
-    if debug:
-        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
-        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
-
-        wanted_params = len(frozen_param_shapes)
-        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
-        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
-        print(f'Frozen params: Have {avail_numel} numels to process.')
-        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
-
-    total_params = 0
-    total_numel = 0
-    for name, shape in frozen_param_shapes.items():
-        total_params += 1
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-
-        state_dict[name] = frozen_param_fragments[name]
-
-        if debug:
-            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
-
-    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _has_callable(obj, fn):
-    attr = getattr(obj, fn, None)
-    return callable(attr)
-
-
-def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
-    param_shapes = zero_model_states[0].param_shapes
-
-    # Reconstruction protocol:
-    #
-    # XXX: document this
-
-    if debug:
-        for i in range(world_size):
-            for j in range(len(fp32_flat_groups[0])):
-                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
-
-    # XXX: memory usage doubles here (zero2)
-    num_param_groups = len(fp32_flat_groups[0])
-    merged_single_partition_of_fp32_groups = []
-    for i in range(num_param_groups):
-        merged_partitions = [sd[i] for sd in fp32_flat_groups]
-        full_single_fp32_vector = torch.cat(merged_partitions, 0)
-        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
-    avail_numel = sum(
-        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
-
-    if debug:
-        wanted_params = sum([len(shapes) for shapes in param_shapes])
-        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
-        # not asserting if there is a mismatch due to possible padding
-        print(f"Have {avail_numel} numels to process.")
-        print(f"Need {wanted_numel} numels in {wanted_params} params.")
-
-    # params
-    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
-    # out-of-core computing solution
-    total_numel = 0
-    total_params = 0
-    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
-        offset = 0
-        avail_numel = full_single_fp32_vector.numel()
-        for name, shape in shapes.items():
-
-            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
-            total_numel += unpartitioned_numel
-            total_params += 1
-
-            if debug:
-                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
-            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
-            offset += unpartitioned_numel
-
-        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
-        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
-        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
-        # live optimizer object, so we are checking that the numbers are within the right range
-        align_to = 2 * world_size
-
-        def zero2_align(x):
-            return align_to * math.ceil(x / align_to)
-
-        if debug:
-            print(f"original offset={offset}, avail_numel={avail_numel}")
-
-        offset = zero2_align(offset)
-        avail_numel = zero2_align(avail_numel)
-
-        if debug:
-            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
-
-        # Sanity check
-        if offset != avail_numel:
-            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
-
-    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                               exclude_frozen_parameters):
-    state_dict = OrderedDict()
-
-    # buffers
-    buffers = zero_model_states[0].buffers
-    state_dict.update(buffers)
-    if debug:
-        print(f"added {len(buffers)} buffers")
-
-    if not exclude_frozen_parameters:
-        _zero2_merge_frozen_params(state_dict, zero_model_states)
-
-    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
-
-    # recover shared parameters
-    for pair in zero_model_states[0].shared_params:
-        if pair[1] in state_dict:
-            state_dict[pair[0]] = state_dict[pair[1]]
-
-    return state_dict
-
-
-def zero3_partitioned_param_info(unpartitioned_numel, world_size):
-    remainder = unpartitioned_numel % world_size
-    padding_numel = (world_size - remainder) if remainder else 0
-    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
-    return partitioned_numel, padding_numel
-
-
-def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
-    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
-        return
-
-    if debug:
-        for i in range(world_size):
-            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
-            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
-
-        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
-        wanted_params = len(frozen_param_shapes)
-        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
-        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
-        print(f'Frozen params: Have {avail_numel} numels to process.')
-        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
-
-    total_params = 0
-    total_numel = 0
-    for name, shape in zero_model_states[0].frozen_param_shapes.items():
-        total_params += 1
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-
-        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
-        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
-
-        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
-
-        if debug:
-            print(
-                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
-            )
-
-    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
-
-
-class GatheredTensor:
-    """
-    A pseudo tensor that collects partitioned weights.
-    It is more memory efficient when there are multiple groups.
-    """
-
-    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
-        self.flat_groups = flat_groups
-        self.flat_groups_offset = flat_groups_offset
-        self.offset = offset
-        self.partitioned_numel = partitioned_numel
-        self.shape = shape
-        self.dtype = self.flat_groups[0][0].dtype
-
-    def contiguous(self):
-        """
-        Merge partitioned weights from flat_groups into a single tensor.
-        """
-        end_idx = self.offset + self.partitioned_numel
-        world_size = len(self.flat_groups)
-        pad_flat_param_chunks = []
-
-        for rank_i in range(world_size):
-            # for each rank, we need to collect weights from related group/groups
-            flat_groups_at_rank_i = self.flat_groups[rank_i]
-            start_group_id = None
-            end_group_id = None
-            for group_id in range(len(self.flat_groups_offset)):
-                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
-                    start_group_id = group_id
-                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
-                    end_group_id = group_id
-                    break
-            # collect weights from related group/groups
-            for group_id in range(start_group_id, end_group_id + 1):
-                flat_tensor = flat_groups_at_rank_i[group_id]
-                start_offset = self.offset - self.flat_groups_offset[group_id]
-                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
-                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
-
-        # collect weights from all ranks
-        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
-        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
-        return param
-
-
-def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
-    param_shapes = zero_model_states[0].param_shapes
-    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
-
-    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
-    # param, re-consolidating each param, while dealing with padding if any
-
-    # merge list of dicts, preserving order
-    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
-
-    if debug:
-        for i in range(world_size):
-            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
-
-        wanted_params = len(param_shapes)
-        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
-        # not asserting if there is a mismatch due to possible padding
-        avail_numel = fp32_flat_groups[0].numel() * world_size
-        print(f"Trainable params: Have {avail_numel} numels to process.")
-        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
-
-    # params
-    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
-    # out-of-core computing solution
-    offset = 0
-    total_numel = 0
-    total_params = 0
-    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
-    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-        total_params += 1
-        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
-
-        if debug:
-            print(
-                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
-            )
-
-        # memory efficient tensor
-        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
-        state_dict[name] = tensor
-        offset += partitioned_numel
-
-    offset *= world_size
-
-    # Sanity check
-    if offset != avail_numel:
-        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
-
-    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                               exclude_frozen_parameters):
-    state_dict = OrderedDict()
-
-    # buffers
-    buffers = zero_model_states[0].buffers
-    state_dict.update(buffers)
-    if debug:
-        print(f"added {len(buffers)} buffers")
-
-    if not exclude_frozen_parameters:
-        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
-
-    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
-
-    # recover shared parameters
-    for pair in zero_model_states[0].shared_params:
-        if pair[1] in state_dict:
-            state_dict[pair[0]] = state_dict[pair[1]]
-
-    return state_dict
-
-
-def to_torch_tensor(state_dict, return_empty_tensor=False):
-    """
-    Convert state_dict of GatheredTensor to torch tensor
-    """
-    torch_state_dict = {}
-    converted_tensors = {}
-    for name, tensor in state_dict.items():
-        tensor_id = id(tensor)
-        if tensor_id in converted_tensors:  # shared tensors
-            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
-            torch_state_dict[name] = shared_tensor
-        else:
-            converted_tensors[tensor_id] = name
-            if return_empty_tensor:
-                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
-            else:
-                torch_state_dict[name] = tensor.contiguous()
-    return torch_state_dict
-
-
-def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
-                                             tag=None,
-                                             exclude_frozen_parameters=False,
-                                             lazy_mode=False):
-    """
-    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
-    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
-    via a model hub.
-
-    Args:
-        - ``checkpoint_dir``: path to the desired checkpoint folder
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
-        - ``exclude_frozen_parameters``: exclude frozen parameters
-        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
-          Convert the pesduo tensor to torch tensor by ``.contiguous()``
-
-    Returns:
-        - pytorch ``state_dict``
-
-    A typical usage might be ::
-
-        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
-        # do the training and checkpoint saving
-        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
-        model = model.cpu() # move to cpu
-        model.load_state_dict(state_dict)
-        # submit to model hub or save the model to share with others
-
-    In this example the ``model`` will no longer be usable in the deepspeed context of the same
-    application. i.e. you will need to re-initialize the deepspeed engine, since
-    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
-
-    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
-
-    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
-    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
-    the checkpoint. Or you can load state_dict in lazy mode ::
-
-        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
-        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
-        for name, lazy_tensor in state_dict.item():
-            tensor = lazy_tensor.contiguous()  # to cpu
-            print(name, tensor)
-            # del tensor to release memory if it no longer in use
-    """
-    if tag is None:
-        latest_path = os.path.join(checkpoint_dir, 'latest')
-        if os.path.isfile(latest_path):
-            with open(latest_path, 'r') as fd:
-                tag = fd.read().strip()
-        else:
-            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
-
-    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
-
-    if not os.path.isdir(ds_checkpoint_dir):
-        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
-
-    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
-    if lazy_mode:
-        return state_dict
-    else:
-        return to_torch_tensor(state_dict)
-
-
-def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
-                                               output_dir,
-                                               max_shard_size="5GB",
-                                               safe_serialization=False,
-                                               tag=None,
-                                               exclude_frozen_parameters=False):
-    """
-    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
-    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
-
-    Args:
-        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
-        - ``output_dir``: directory to the pytorch fp32 state_dict output files
-        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
-        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
-        - ``exclude_frozen_parameters``: exclude frozen parameters
-    """
-
-    # Dependency pre-check
-    if safe_serialization:
-        try:
-            from safetensors.torch import save_file
-        except ImportError:
-            print('If you want to use `safe_serialization`, please `pip install safetensors`')
-            raise
-    if max_shard_size is not None:
-        try:
-            from huggingface_hub import split_torch_state_dict_into_shards
-        except ImportError:
-            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
-            raise
-
-    # Convert zero checkpoint to state_dict
-    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
-                                                          tag,
-                                                          exclude_frozen_parameters,
-                                                          lazy_mode=True)
-
-    # Shard the model if it is too big.
-    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
-    if max_shard_size is not None:
-        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
-        # an memory-efficient approach for sharding
-        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
-        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
-                                                              filename_pattern=filename_pattern,
-                                                              max_shard_size=max_shard_size)
-    else:
-        from collections import namedtuple
-        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
-        state_dict_split = StateDictSplit(is_sharded=False,
-                                          filename_to_tensors={weights_name: list(state_dict.keys())})
-
-    # Save the model by shard
-    os.makedirs(output_dir, exist_ok=True)
-    filename_to_tensors = state_dict_split.filename_to_tensors.items()
-    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
-        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
-        shard_state_dict = to_torch_tensor(shard_state_dict)
-        output_path = os.path.join(output_dir, shard_file)
-        if safe_serialization:
-            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
-        else:
-            torch.save(shard_state_dict, output_path)
-        # release the memory of current shard
-        for tensor_name in list(shard_state_dict.keys()):
-            del state_dict[tensor_name]
-            del shard_state_dict[tensor_name]
-        del shard_state_dict
-        gc.collect()
-
-    # Save index if sharded
-    if state_dict_split.is_sharded:
-        index = {
-            "metadata": state_dict_split.metadata,
-            "weight_map": state_dict_split.tensor_to_filename,
-        }
-        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
-        save_index_file = os.path.join(output_dir, save_index_file)
-        with open(save_index_file, "w", encoding="utf-8") as f:
-            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
-            f.write(content)
-
-
-def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
-    """
-    1. Put the provided model to cpu
-    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
-    3. Load it into the provided model
-
-    Args:
-        - ``model``: the model object to update
-        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
-
-    Returns:
-        - ``model`: modified model
-
-    Make sure you have plenty of CPU memory available before you call this function. If you don't
-    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
-    conveniently placed for you in the checkpoint folder.
-
-    A typical usage might be ::
-
-        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
-        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
-        # submit to model hub or save the model to share with others
-
-    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
-    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
-    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
-
-    """
-    logger.info("Extracting fp32 weights")
-    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
-
-    logger.info("Overwriting model with fp32 weights")
-    model = model.cpu()
-    model.load_state_dict(state_dict, strict=False)
-
-    return model
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument("checkpoint_dir",
-                        type=str,
-                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
-    parser.add_argument("output_dir",
-                        type=str,
-                        help="directory to the pytorch fp32 state_dict output files"
-                        "(e.g. path/checkpoint-12-output/)")
-    parser.add_argument(
-        "--max_shard_size",
-        type=str,
-        default="5GB",
-        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
-        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
-        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
-        "without CPU OOM issues.")
-    parser.add_argument(
-        "--safe_serialization",
-        default=False,
-        action='store_true',
-        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
-    parser.add_argument("-t",
-                        "--tag",
-                        type=str,
-                        default=None,
-                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
-    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
-    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
-    args = parser.parse_args()
-
-    debug = args.debug
-
-    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
-                                               args.output_dir,
-                                               max_shard_size=args.max_shard_size,
-                                               safe_serialization=args.safe_serialization,
-                                               tag=args.tag,
-                                               exclude_frozen_parameters=args.exclude_frozen_parameters)
diff --git a/ckpts/v10_2_32f/args.json b/ckpts/v10_2_32f/args.json
deleted file mode 100644
index 55005790271a0106fa4d692419406ff250593e12..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/args.json
+++ /dev/null
@@ -1,376 +0,0 @@
-{
-  "output_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S2/4b/qwen3_5_2b/v10_2_32f_10pct/v0-20260610-154054",
-  "per_device_train_batch_size": 1,
-  "num_train_epochs": 1.0,
-  "max_steps": -1,
-  "learning_rate": 1e-05,
-  "lr_scheduler_type": "cosine",
-  "lr_scheduler_kwargs": null,
-  "warmup_steps": 0,
-  "optim": "adamw_torch_fused",
-  "optim_args": null,
-  "weight_decay": 0.1,
-  "adam_beta1": 0.9,
-  "adam_beta2": 0.95,
-  "adam_epsilon": 1e-08,
-  "optim_target_modules": null,
-  "gradient_accumulation_steps": 8,
-  "average_tokens_across_devices": true,
-  "max_grad_norm": 1.0,
-  "label_smoothing_factor": 0.0,
-  "bf16": true,
-  "fp16": false,
-  "bf16_full_eval": false,
-  "fp16_full_eval": false,
-  "tf32": null,
-  "gradient_checkpointing": true,
-  "gradient_checkpointing_kwargs": "{\"use_reentrant\": false}",
-  "torch_compile": false,
-  "torch_compile_backend": null,
-  "torch_compile_mode": null,
-  "use_liger_kernel": false,
-  "liger_kernel_config": null,
-  "use_cache": false,
-  "neftune_noise_alpha": null,
-  "torch_empty_cache_steps": null,
-  "auto_find_batch_size": false,
-  "logging_strategy": "steps",
-  "logging_steps": 1,
-  "logging_first_step": true,
-  "log_on_each_node": true,
-  "logging_nan_inf_filter": true,
-  "include_num_input_tokens_seen": false,
-  "log_level": "passive",
-  "log_level_replica": "warning",
-  "disable_tqdm": null,
-  "report_to": [
-    "none"
-  ],
-  "run_name": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S2/4b/qwen3_5_2b/v10_2_32f_10pct/v0-20260610-154054",
-  "project": "huggingface",
-  "trackio_space_id": "trackio",
-  "eval_strategy": "no",
-  "eval_steps": 200.0,
-  "eval_delay": 0,
-  "per_device_eval_batch_size": 1,
-  "prediction_loss_only": false,
-  "eval_on_start": false,
-  "eval_do_concat_batches": true,
-  "eval_use_gather_object": false,
-  "eval_accumulation_steps": null,
-  "include_for_metrics": [],
-  "batch_eval_metrics": false,
-  "save_only_model": false,
-  "save_strategy": "steps",
-  "save_steps": 200.0,
-  "save_on_each_node": false,
-  "save_total_limit": 1,
-  "enable_jit_checkpoint": false,
-  "push_to_hub": false,
-  "hub_token": null,
-  "hub_private_repo": null,
-  "hub_model_id": null,
-  "hub_strategy": "every_save",
-  "hub_always_push": false,
-  "hub_revision": null,
-  "load_best_model_at_end": false,
-  "metric_for_best_model": "loss",
-  "greater_is_better": false,
-  "ignore_data_skip": false,
-  "restore_callback_states_from_checkpoint": false,
-  "full_determinism": false,
-  "seed": 42,
-  "data_seed": 42,
-  "use_cpu": false,
-  "accelerator_config": "{\"dispatch_batches\": false}",
-  "parallelism_config": null,
-  "dataloader_drop_last": false,
-  "dataloader_num_workers": 6,
-  "dataloader_pin_memory": true,
-  "dataloader_persistent_workers": true,
-  "dataloader_prefetch_factor": 4,
-  "remove_unused_columns": true,
-  "label_names": null,
-  "train_sampling_strategy": "random",
-  "length_column_name": "length",
-  "ddp_find_unused_parameters": null,
-  "ddp_bucket_cap_mb": null,
-  "ddp_broadcast_buffers": null,
-  "ddp_backend": null,
-  "ddp_timeout": 7200,
-  "fsdp": [],
-  "fsdp_config": null,
-  "deepspeed": {
-    "fp16": {
-      "enabled": "auto",
-      "loss_scale": 0,
-      "loss_scale_window": 1000,
-      "initial_scale_power": 16,
-      "hysteresis": 2,
-      "min_loss_scale": 1
-    },
-    "bf16": {
-      "enabled": "auto"
-    },
-    "zero_optimization": {
-      "stage": 2,
-      "offload_optimizer": {
-        "device": "none",
-        "pin_memory": true
-      },
-      "allgather_partitions": true,
-      "allgather_bucket_size": 200000000.0,
-      "overlap_comm": false,
-      "reduce_scatter": true,
-      "reduce_bucket_size": 200000000.0,
-      "contiguous_gradients": true
-    },
-    "gradient_accumulation_steps": "auto",
-    "gradient_clipping": "auto",
-    "steps_per_print": 2000,
-    "train_batch_size": "auto",
-    "train_micro_batch_size_per_gpu": "auto",
-    "wall_clock_breakdown": false
-  },
-  "debug": null,
-  "skip_memory_metrics": true,
-  "do_train": false,
-  "do_eval": false,
-  "do_predict": false,
-  "resume_from_checkpoint": null,
-  "warmup_ratio": 0.05,
-  "logging_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S2/4b/qwen3_5_2b/v10_2_32f_10pct/v0-20260610-154054/runs",
-  "local_rank": 0,
-  "sortish_sampler": false,
-  "predict_with_generate": false,
-  "generation_max_length": null,
-  "generation_num_beams": null,
-  "generation_config": null,
-  "tuner_backend": "peft",
-  "vit_gradient_checkpointing": true,
-  "router_aux_loss_coef": 0.0,
-  "enable_dft_loss": false,
-  "enable_channel_loss": false,
-  "safe_serialization": true,
-  "max_shard_size": "5GB",
-  "check_model": true,
-  "acc_strategy": "token",
-  "train_dataloader_shuffle": true,
-  "group_by_length": false,
-  "max_epochs": null,
-  "aligner_lr": null,
-  "vit_lr": 1e-06,
-  "use_logits_to_keep": null,
-  "ds3_gather_for_generation": true,
-  "resume_only_model": false,
-  "optimizer": null,
-  "loss_type": null,
-  "eval_metric": null,
-  "callbacks": [],
-  "early_stop_interval": null,
-  "eval_use_evalscope": false,
-  "eval_dataset": [],
-  "eval_dataset_args": null,
-  "eval_limit": null,
-  "eval_generation_config": null,
-  "extra_eval_args": null,
-  "tuner_type": "full",
-  "use_galore": false,
-  "galore_target_modules": null,
-  "galore_rank": 128,
-  "galore_update_proj_gap": 50,
-  "galore_scale": 1.0,
-  "galore_proj_type": "std",
-  "galore_optim_per_parameter": false,
-  "galore_with_embedding": false,
-  "galore_quantization": false,
-  "galore_proj_quant": false,
-  "galore_proj_bits": 4,
-  "galore_proj_group_size": 256,
-  "galore_cos_threshold": 0.4,
-  "galore_gamma_proj": 2,
-  "galore_queue_size": 5,
-  "lisa_activated_layers": 0,
-  "lisa_step_interval": 20,
-  "use_flash_ckpt": false,
-  "use_ray": false,
-  "ray_exp_name": null,
-  "device_groups": null,
-  "model": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S1/4b/qwen3_5_2b/v10_2/v0-20260610-091722/checkpoint-2181",
-  "model_type": "llava_qwen3_5vit_qwen3",
-  "model_revision": null,
-  "task_type": "causal_lm",
-  "torch_dtype": "bfloat16",
-  "attn_impl": "flash_attn",
-  "experts_impl": null,
-  "new_special_tokens": [],
-  "num_labels": null,
-  "problem_type": null,
-  "rope_scaling": null,
-  "device_map": null,
-  "max_memory": {},
-  "max_model_len": null,
-  "local_repo_path": null,
-  "init_strategy": null,
-  "template": "llava_qwen3_5vit_qwen3",
-  "system": null,
-  "max_length": 16384,
-  "truncation_strategy": "delete",
-  "max_pixels": null,
-  "agent_template": null,
-  "norm_bbox": null,
-  "use_chat_template": true,
-  "padding_side": "right",
-  "padding_free": true,
-  "loss_scale": "default",
-  "sequence_parallel_size": 1,
-  "template_backend": "swift",
-  "response_prefix": null,
-  "enable_thinking": null,
-  "add_non_thinking_prefix": true,
-  "dataset": [],
-  "val_dataset": [],
-  "cached_dataset": [
-    "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/image_10pct/train",
-    "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/video_10pct_32f/train"
-  ],
-  "cached_val_dataset": [],
-  "split_dataset_ratio": 0.0,
-  "dataset_num_proc": 16,
-  "load_from_cache_file": false,
-  "dataset_shuffle": true,
-  "val_dataset_shuffle": false,
-  "streaming": false,
-  "interleave_prob": null,
-  "stopping_strategy": "first_exhausted",
-  "shuffle_buffer_size": 1000,
-  "download_mode": "reuse_dataset_if_exists",
-  "columns": {},
-  "strict": false,
-  "model_name": null,
-  "model_author": null,
-  "custom_dataset_info": [],
-  "quant_method": null,
-  "quant_bits": null,
-  "hqq_axis": null,
-  "bnb_4bit_compute_dtype": "bfloat16",
-  "bnb_4bit_quant_type": "nf4",
-  "bnb_4bit_use_double_quant": true,
-  "bnb_4bit_quant_storage": null,
-  "max_new_tokens": 64,
-  "temperature": 0.0,
-  "top_k": null,
-  "top_p": null,
-  "repetition_penalty": null,
-  "num_beams": 1,
-  "stream": false,
-  "stop_words": [],
-  "logprobs": false,
-  "top_logprobs": null,
-  "structured_outputs_regex": null,
-  "adapters": [],
-  "external_plugins": [
-    "ms-swift/video_mllm/model_plugin_qwen3_5vit.py",
-    "ms-swift/video_mllm/dataset_plugin.py"
-  ],
-  "custom_register_path": [],
-  "model_kwargs": {},
-  "load_args": false,
-  "load_data_args": false,
-  "packing": true,
-  "packing_length": 16384,
-  "packing_num_proc": 1,
-  "lazy_tokenize": false,
-  "use_hf": true,
-  "ignore_args_error": false,
-  "use_swift_lora": false,
-  "freeze_parameters": [],
-  "freeze_parameters_regex": null,
-  "freeze_parameters_ratio": 0.0,
-  "trainable_parameters": [
-    "model.multi_modal_projector"
-  ],
-  "trainable_parameters_regex": null,
-  "freeze_llm": false,
-  "freeze_vit": false,
-  "freeze_aligner": false,
-  "target_modules": [
-    "all-linear"
-  ],
-  "target_regex": null,
-  "target_parameters": null,
-  "modules_to_save": [],
-  "lora_rank": 8,
-  "lora_alpha": 32,
-  "lora_dropout": 0.05,
-  "lora_bias": "none",
-  "lora_dtype": null,
-  "lorap_lr_ratio": null,
-  "use_rslora": false,
-  "use_dora": false,
-  "lora_ga_batch_size": 2,
-  "lora_ga_iters": 2,
-  "lora_ga_max_length": 1024,
-  "lora_ga_direction": "ArB2r",
-  "lora_ga_scale": "stable",
-  "lora_ga_stable_gamma": 16,
-  "init_weights": true,
-  "fourier_n_frequency": 2000,
-  "fourier_scaling": 300.0,
-  "boft_block_size": 4,
-  "boft_block_num": 0,
-  "boft_n_butterfly_factor": 1,
-  "boft_dropout": 0.0,
-  "vera_rank": 256,
-  "vera_projection_prng_key": 0,
-  "vera_dropout": 0.0,
-  "vera_d_initial": 0.1,
-  "adapter_act": "gelu",
-  "adapter_length": 128,
-  "adalora_target_r": 8,
-  "adalora_init_r": 12,
-  "adalora_tinit": 0,
-  "adalora_tfinal": 0,
-  "adalora_deltaT": 1,
-  "adalora_beta1": 0.85,
-  "adalora_beta2": 0.85,
-  "adalora_orth_reg_weight": 0.5,
-  "llamapro_num_new_blocks": 4,
-  "llamapro_num_groups": null,
-  "reft_layer_key": null,
-  "reft_layers": null,
-  "reft_rank": 4,
-  "reft_intervention_type": "LoreftIntervention",
-  "reft_args": null,
-  "swanlab_token": null,
-  "swanlab_project": "ms-swift",
-  "swanlab_workspace": null,
-  "swanlab_exp_name": null,
-  "swanlab_notification_method": null,
-  "swanlab_webhook_url": null,
-  "swanlab_secret": null,
-  "swanlab_sender_email": null,
-  "swanlab_receiver_email": null,
-  "swanlab_smtp_server": null,
-  "swanlab_smtp_port": null,
-  "swanlab_email_language": "zh",
-  "swanlab_mode": "cloud",
-  "add_version": true,
-  "create_checkpoint_symlink": false,
-  "zero_hpz_partition_size": null,
-  "deepspeed_autotp_size": null,
-  "swift_version": "4.1.1",
-  "ckpt_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S1/4b/qwen3_5_2b/v10_2/v0-20260610-091722/checkpoint-2181",
-  "rank": 0,
-  "global_world_size": 8,
-  "local_world_size": 8,
-  "model_suffix": "checkpoint-2181",
-  "model_info": "ModelInfo(model_type='llava_qwen3_5vit_qwen3', model_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S1/4b/qwen3_5_2b/v10_2/v0-20260610-091722/checkpoint-2181', torch_dtype=torch.bfloat16, max_model_len=262144, quant_method=None, quant_bits=None, rope_scaling=None, is_moe_model=False, is_multimodal=True, config=None, task_type='causal_lm', num_labels=None)",
-  "model_meta": "ModelMeta(model_type='llava_qwen3_5vit_qwen3', model_groups=[ModelGroup(models=[Model(ms_model_id='Qwen/Qwen3-0.6B', hf_model_id='Qwen/Qwen3-0.6B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-1.7B', hf_model_id='Qwen/Qwen3-1.7B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-4B', hf_model_id='Qwen/Qwen3-4B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-8B', hf_model_id='Qwen/Qwen3-8B', model_path=None, ms_revision=None, hf_revision=None)], template=None, ignore_patterns=None, requires=None, tags=[])], loader=<class 'model_plugin_qwen3_5vit.Qwen3_5ViTQwen3Loader'>, template='llava_qwen3_5vit_qwen3', model_arch=MultiModelKeys(arch_name='llava_hf', embedding=None, module_list=None, lm_head=None, q_proj=None, k_proj=None, v_proj=None, o_proj=None, attention=None, mlp=None, down_proj=None, qkv_proj=None, qk_proj=None, qa_proj=None, qb_proj=None, kv_proj=None, kva_proj=None, kvb_proj=None, language_model=['model.language_model', 'lm_head'], aligner=['model.multi_modal_projector'], vision_tower=['model.vision_tower'], generator=[]), mcore_model_type=None, architectures=['LlavaQwen3_5ViTForConditionalGeneration'], additional_saved_files=[], torch_dtype=None, is_multimodal=True, is_reward=False, task_type=None, ignore_patterns=None, requires=[], tags=['vision', 'video'])",
-  "model_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S1/4b/qwen3_5_2b/v10_2/v0-20260610-091722/checkpoint-2181",
-  "template_meta": "QwenTemplateMeta(template_type='llava_qwen3_5vit_qwen3', prefix=[], prompt=['<|im_start|>user\\n{{QUERY}}<|im_end|>\\n<|im_start|>assistant\\n'], chat_sep=['<|im_end|>\\n'], suffix=['<|im_end|>\\n'], template_cls=<class 'model_plugin_qwen3_5vit.Qwen3_5ViTLlavaTemplate'>, system_prefix=['<|im_start|>system\\n{{SYSTEM}}<|im_end|>\\n'], default_system=None, auto_add_bos=False, stop_words=['<|endoftext|>'], agent_template='hermes', is_thinking=False, thinking_prefix='', non_thinking_prefix='', history_thinking_prefix='')",
-  "_val_dataset_exists": false,
-  "hub": "<class 'swift.hub.hub.HFHub'>",
-  "evaluation_strategy": "steps",
-  "training_args": "Seq2SeqTrainingArguments(output_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S2/4b/qwen3_5_2b/v10_2_32f_10pct/v0-20260610-154054', per_device_train_batch_size=1, num_train_epochs=1.0, max_steps=-1, learning_rate=1e-05, lr_scheduler_type=<SchedulerType.COSINE: 'cosine'>, lr_scheduler_kwargs=None, warmup_steps=0.05, optim=<OptimizerNames.ADAMW_TORCH_FUSED: 'adamw_torch_fused'>, optim_args=None, weight_decay=0.1, adam_beta1=0.9, adam_beta2=0.95, adam_epsilon=1e-08, optim_target_modules=None, gradient_accumulation_steps=8, average_tokens_across_devices=None, max_grad_norm=1.0, label_smoothing_factor=0.0, bf16=True, fp16=False, bf16_full_eval=False, fp16_full_eval=False, tf32=None, gradient_checkpointing=True, gradient_checkpointing_kwargs={'use_reentrant': False}, torch_compile=False, torch_compile_backend=None, torch_compile_mode=None, use_liger_kernel=False, liger_kernel_config=None, use_cache=False, neftune_noise_alpha=None, torch_empty_cache_steps=None, auto_find_batch_size=False, logging_strategy=<IntervalStrategy.STEPS: 'steps'>, logging_steps=1, logging_first_step=True, log_on_each_node=True, logging_nan_inf_filter=True, include_num_input_tokens_seen=None, log_level='passive', log_level_replica='warning', disable_tqdm=False, report_to=[], run_name='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S2/4b/qwen3_5_2b/v10_2_32f_10pct/v0-20260610-154054', project='huggingface', trackio_space_id='trackio', eval_strategy=<IntervalStrategy.NO: 'no'>, eval_steps=200.0, eval_delay=0, per_device_eval_batch_size=1, prediction_loss_only=False, eval_on_start=False, eval_do_concat_batches=True, eval_use_gather_object=False, eval_accumulation_steps=None, include_for_metrics=[], batch_eval_metrics=False, save_only_model=False, save_strategy=<SaveStrategy.STEPS: 'steps'>, save_steps=200, save_on_each_node=False, save_total_limit=1, enable_jit_checkpoint=False, push_to_hub=False, hub_token=None, hub_private_repo=None, hub_model_id=None, hub_strategy=<HubStrategy.EVERY_SAVE: 'every_save'>, hub_always_push=False, hub_revision=None, load_best_model_at_end=False, metric_for_best_model='loss', greater_is_better=False, ignore_data_skip=False, restore_callback_states_from_checkpoint=False, full_determinism=False, seed=42, data_seed=42, use_cpu=False, accelerator_config=AcceleratorConfig(split_batches=False, dispatch_batches=False, even_batches=True, use_seedable_sampler=True, non_blocking=False, gradient_accumulation_kwargs=None, use_configured_state=False), parallelism_config=None, dataloader_drop_last=False, dataloader_num_workers=6, dataloader_pin_memory=True, dataloader_persistent_workers=True, dataloader_prefetch_factor=4, remove_unused_columns=False, label_names=None, train_sampling_strategy='random', length_column_name='length', ddp_find_unused_parameters=None, ddp_bucket_cap_mb=None, ddp_broadcast_buffers=None, ddp_backend=None, ddp_timeout=7200, fsdp=[], fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}, deepspeed={'fp16': {'enabled': 'auto', 'loss_scale': 0, 'loss_scale_window': 1000, 'initial_scale_power': 16, 'hysteresis': 2, 'min_loss_scale': 1}, 'bf16': {'enabled': 'auto'}, 'zero_optimization': {'stage': 2, 'offload_optimizer': {'device': 'none', 'pin_memory': True}, 'allgather_partitions': True, 'allgather_bucket_size': 200000000.0, 'overlap_comm': False, 'reduce_scatter': True, 'reduce_bucket_size': 200000000.0, 'contiguous_gradients': True}, 'gradient_accumulation_steps': 'auto', 'gradient_clipping': 'auto', 'steps_per_print': 2000, 'train_batch_size': 'auto', 'train_micro_batch_size_per_gpu': 'auto', 'wall_clock_breakdown': False}, debug=[], skip_memory_metrics=True, do_train=False, do_eval=False, do_predict=False, resume_from_checkpoint=None, warmup_ratio=0.05, logging_dir='/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/S2/4b/qwen3_5_2b/v10_2_32f_10pct/v0-20260610-154054/runs', local_rank=0, sortish_sampler=False, predict_with_generate=False, generation_max_length=None, generation_num_beams=None, generation_config=None, tuner_backend='peft', vit_gradient_checkpointing=True, router_aux_loss_coef=0.0, enable_dft_loss=False, enable_channel_loss=False, safe_serialization=True, max_shard_size='5GB', check_model=True, acc_strategy='token', train_dataloader_shuffle=True, group_by_length=False, max_epochs=None, aligner_lr=None, vit_lr=1e-06, use_logits_to_keep=None, ds3_gather_for_generation=True, resume_only_model=False, optimizer='multimodal', loss_type=None, eval_metric=None, callbacks=[], early_stop_interval=None, eval_use_evalscope=False, eval_dataset=[], eval_dataset_args=None, eval_limit=None, eval_generation_config=None, extra_eval_args=None, tuner_type='full', use_galore=False, galore_target_modules=None, galore_rank=128, galore_update_proj_gap=50, galore_scale=1.0, galore_proj_type='std', galore_optim_per_parameter=False, galore_with_embedding=False, galore_quantization=False, galore_proj_quant=False, galore_proj_bits=4, galore_proj_group_size=256, galore_cos_threshold=0.4, galore_gamma_proj=2, galore_queue_size=5, lisa_activated_layers=0, lisa_step_interval=20, use_flash_ckpt=False)"
-}
\ No newline at end of file
diff --git a/ckpts/v10_2_32f/chat_template.jinja b/ckpts/v10_2_32f/chat_template.jinja
deleted file mode 100644
index 70adff8a08fb31e0636f618564838d4bf3c05286..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/chat_template.jinja
+++ /dev/null
@@ -1,61 +0,0 @@
-{%- if tools %}
-    {{- '<|im_start|>system\n' }}
-    {%- if messages[0].role == 'system' %}
-        {{- messages[0].content + '\n\n' }}
-    {%- endif %}
-    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
-    {%- for tool in tools %}
-        {{- "\n" }}
-        {{- tool | tojson }}
-    {%- endfor %}
-    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
-{%- else %}
-    {%- if messages[0].role == 'system' %}
-        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
-    {%- endif %}
-{%- endif %}
-{%- for message in messages %}
-    {%- if message.content is string %}
-        {%- set content = message.content %}
-    {%- else %}
-        {%- set content = '' %}
-    {%- endif %}
-    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
-        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
-    {%- elif message.role == "assistant" %}
-        {{- '<|im_start|>' + message.role + '\n' + content }}
-        {%- if message.tool_calls %}
-            {%- for tool_call in message.tool_calls %}
-                {%- if (loop.first and content) or (not loop.first) %}
-                    {{- '\n' }}
-                {%- endif %}
-                {%- if tool_call.function %}
-                    {%- set tool_call = tool_call.function %}
-                {%- endif %}
-                {{- '<tool_call>\n{"name": "' }}
-                {{- tool_call.name }}
-                {{- '", "arguments": ' }}
-                {%- if tool_call.arguments is string %}
-                    {{- tool_call.arguments }}
-                {%- else %}
-                    {{- tool_call.arguments | tojson }}
-                {%- endif %}
-                {{- '}\n</tool_call>' }}
-            {%- endfor %}
-        {%- endif %}
-        {{- '<|im_end|>\n' }}
-    {%- elif message.role == "tool" %}
-        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
-            {{- '<|im_start|>user' }}
-        {%- endif %}
-        {{- '\n<tool_response>\n' }}
-        {{- content }}
-        {{- '\n</tool_response>' }}
-        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
-            {{- '<|im_end|>\n' }}
-        {%- endif %}
-    {%- endif %}
-{%- endfor %}
-{%- if add_generation_prompt %}
-    {{- '<|im_start|>assistant\n' }}
-{%- endif %}
\ No newline at end of file
diff --git a/ckpts/v10_2_32f/chat_template.json b/ckpts/v10_2_32f/chat_template.json
deleted file mode 100644
index 68724c68b1612e9ab2e31f84e3279099ad656ff6..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/chat_template.json
+++ /dev/null
@@ -1,3 +0,0 @@
-{
-  "chat_template": "{% for message in messages %}{{'<|im_start|>' + message['role'] + ' '}}{# Render all images first #}{% for content in message['content'] | selectattr('type', 'equalto', 'image') %}{{ '<image>' }}{% endfor %}{# Render all video then #}{% for content in message['content'] | selectattr('type', 'equalto', 'video') %}{{ '<video>' }}{% endfor %}{# Render all text next #}{% if message['role'] != 'assistant' %}{% for content in message['content'] | selectattr('type', 'equalto', 'text') %}{{ '\n' + content['text'] }}{% endfor %}{% else %}{% for content in message['content'] | selectattr('type', 'equalto', 'text') %}{% generation %}{{ '\n' + content['text'] }}{% endgeneration %}{% endfor %}{% endif %}{{'<|im_end|>'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"
-}
diff --git a/ckpts/v10_2_32f/config.json b/ckpts/v10_2_32f/config.json
deleted file mode 100644
index 7ea259f457ed29a71d50c17eb0da990fcf98924a..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/config.json
+++ /dev/null
@@ -1,264 +0,0 @@
-{
-  "architectures": [
-    "LlavaQwen3_5ViTForConditionalGeneration"
-  ],
-  "bos_token_id": null,
-  "dtype": "bfloat16",
-  "eos_token_id": 151645,
-  "hidden_size": 2560,
-  "image_grid_pinpoints": [
-    [
-      384,
-      384
-    ],
-    [
-      384,
-      768
-    ],
-    [
-      384,
-      1152
-    ],
-    [
-      384,
-      1536
-    ],
-    [
-      384,
-      1920
-    ],
-    [
-      384,
-      2304
-    ],
-    [
-      768,
-      384
-    ],
-    [
-      768,
-      768
-    ],
-    [
-      768,
-      1152
-    ],
-    [
-      768,
-      1536
-    ],
-    [
-      768,
-      1920
-    ],
-    [
-      768,
-      2304
-    ],
-    [
-      1152,
-      384
-    ],
-    [
-      1152,
-      768
-    ],
-    [
-      1152,
-      1152
-    ],
-    [
-      1152,
-      1536
-    ],
-    [
-      1152,
-      1920
-    ],
-    [
-      1152,
-      2304
-    ],
-    [
-      1536,
-      384
-    ],
-    [
-      1536,
-      768
-    ],
-    [
-      1536,
-      1152
-    ],
-    [
-      1536,
-      1536
-    ],
-    [
-      1536,
-      1920
-    ],
-    [
-      1536,
-      2304
-    ],
-    [
-      1920,
-      384
-    ],
-    [
-      1920,
-      768
-    ],
-    [
-      1920,
-      1152
-    ],
-    [
-      1920,
-      1536
-    ],
-    [
-      1920,
-      1920
-    ],
-    [
-      1920,
-      2304
-    ],
-    [
-      2304,
-      384
-    ],
-    [
-      2304,
-      768
-    ],
-    [
-      2304,
-      1152
-    ],
-    [
-      2304,
-      1536
-    ],
-    [
-      2304,
-      1920
-    ],
-    [
-      2304,
-      2304
-    ]
-  ],
-  "image_token_index": 151669,
-  "model_type": "llava_qwen3_5vit_qwen3",
-  "multimodal_projector_bias": true,
-  "pad_token_id": 151643,
-  "projector_hidden_act": "gelu",
-  "text_config": {
-    "_name_or_path": "/share/m2v_intern_v3/wangjunjie09/model_cache/huggingface/Qwen/Qwen3-4B-Instruct-2507",
-    "architectures": [
-      "Qwen3ForCausalLM"
-    ],
-    "attention_bias": false,
-    "attention_dropout": 0.0,
-    "bos_token_id": 151643,
-    "dtype": "bfloat16",
-    "eos_token_id": 151645,
-    "head_dim": 128,
-    "hidden_act": "silu",
-    "hidden_size": 2560,
-    "initializer_range": 0.02,
-    "intermediate_size": 9728,
-    "layer_types": [
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention",
-      "full_attention"
-    ],
-    "max_position_embeddings": 262144,
-    "max_window_layers": 36,
-    "model_type": "qwen3",
-    "num_attention_heads": 32,
-    "num_hidden_layers": 36,
-    "num_key_value_heads": 8,
-    "pad_token_id": 151643,
-    "rms_norm_eps": 1e-06,
-    "rope_parameters": {
-      "rope_theta": 5000000,
-      "rope_type": "default"
-    },
-    "sliding_window": null,
-    "tie_word_embeddings": true,
-    "use_cache": false,
-    "use_sliding_window": false,
-    "vocab_size": 151936
-  },
-  "tie_word_embeddings": true,
-  "transformers_version": "5.5.4",
-  "use_cache": false,
-  "video_token_index": 151670,
-  "vision_aspect_ratio": "anyres_max_9",
-  "vision_config": {
-    "deepstack_visual_indexes": [],
-    "depth": 24,
-    "dtype": "bfloat16",
-    "hidden_act": "gelu_pytorch_tanh",
-    "hidden_size": 1024,
-    "image_size": 384,
-    "in_channels": 3,
-    "initializer_range": 0.02,
-    "intermediate_size": 4096,
-    "model_type": "qwen3_5",
-    "num_heads": 16,
-    "num_position_embeddings": 2304,
-    "out_hidden_size": 2048,
-    "patch_size": 16,
-    "spatial_merge_size": 2,
-    "temporal_patch_size": 2
-  },
-  "vision_feature_layer": -1,
-  "vision_feature_select_strategy": "full",
-  "vit_register_ring": 0,
-  "vit_skip_last_blocks": 0,
-  "auto_map": {
-    "AutoConfig": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTConfig",
-    "AutoModel": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
-    "AutoModelForCausalLM": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
-    "AutoModelForImageTextToText": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration"
-  }
-}
\ No newline at end of file
diff --git a/ckpts/v10_2_32f/generation_config.json b/ckpts/v10_2_32f/generation_config.json
deleted file mode 100644
index 56320d3179df3512d0cb76c240831a10e6f7c6f7..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/generation_config.json
+++ /dev/null
@@ -1,12 +0,0 @@
-{
-  "_from_model_config": true,
-  "bos_token_id": 151643,
-  "eos_token_id": [
-    151645,
-    151643
-  ],
-  "output_attentions": false,
-  "output_hidden_states": false,
-  "transformers_version": "5.5.4",
-  "use_cache": true
-}
diff --git a/ckpts/v10_2_32f/latest b/ckpts/v10_2_32f/latest
deleted file mode 100644
index 2c429357c7f2fba84efb27b0672c454288951b67..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/latest
+++ /dev/null
@@ -1 +0,0 @@
-global_step785
\ No newline at end of file
diff --git a/ckpts/v10_2_32f/model-00001-of-00002.safetensors b/ckpts/v10_2_32f/model-00001-of-00002.safetensors
deleted file mode 100644
index 3679bc82cb6ac93ab18240539c7d0ed9b26db134..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/model-00001-of-00002.safetensors
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:128b40150f511cf7ac1cf32f327bf607d867c90870071102cb290005fd309114
-size 4987507408
diff --git a/ckpts/v10_2_32f/model-00002-of-00002.safetensors b/ckpts/v10_2_32f/model-00002-of-00002.safetensors
deleted file mode 100644
index f423c9a833b903349691e10fcb2cae120da95d23..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/model-00002-of-00002.safetensors
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:4ad7114700d79a82f40f2197414ad53e9d50eea8c92c912adb14fe7a72adfc76
-size 4466285248
diff --git a/ckpts/v10_2_32f/model.safetensors.index.json b/ckpts/v10_2_32f/model.safetensors.index.json
deleted file mode 100644
index a227833b5ba2c2ae9deeddde8d4fc716babffc8c..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/model.safetensors.index.json
+++ /dev/null
@@ -1,705 +0,0 @@
-{
-  "metadata": {
-    "total_parameters": 4337895424,
-    "total_size": 9453703168
-  },
-  "weight_map": {
-    "lm_head.weight": "model-00001-of-00002.safetensors",
-    "model.image_newline": "model-00001-of-00002.safetensors",
-    "model.language_model.embed_tokens.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.17.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.18.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.19.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.20.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.35.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
-    "model.language_model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
-    "model.language_model.norm.weight": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_1.bias": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_1.weight": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_2.bias": "model-00002-of-00002.safetensors",
-    "model.multi_modal_projector.linear_2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.0.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.1.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.10.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.11.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.12.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.13.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.14.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.15.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.16.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.17.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.18.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.19.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.2.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.20.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.21.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.22.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.23.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.3.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.4.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.5.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.6.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.7.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.8.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.qkv.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.attn.qkv.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm1.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm1.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm2.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.blocks.9.norm2.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.final_layernorm.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.final_layernorm.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.patch_embed.proj.bias": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.patch_embed.proj.weight": "model-00002-of-00002.safetensors",
-    "model.vision_tower.vision.pos_embed.weight": "model-00002-of-00002.safetensors"
-  }
-}
diff --git a/ckpts/v10_2_32f/modeling_qwen3_5vit_qwen3.py b/ckpts/v10_2_32f/modeling_qwen3_5vit_qwen3.py
deleted file mode 100644
index 05d150ac147b78e77aff902bda1daf0c26759541..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/modeling_qwen3_5vit_qwen3.py
+++ /dev/null
@@ -1,405 +0,0 @@
-"""
-modeling_qwen3_5vit_qwen3.py — Qwen3.5 Vision as SigLIP-compat vision_tower in LlavaOnevision.
-
-设计：与 `modeling_qwen3vlvit_qwen3.py` 严格同构，仅换 vision backbone 源：
-  - Qwen3VLVisionModel → Qwen3_5VisionModel（继承关系：Qwen3_5VisionModel(Qwen3VLVisionModel) 去 DeepStack）
-  - Qwen3VLVisionConfig → Qwen3_5VisionConfig（父类用 AttributeError 哨兵屏蔽 deepstack_visual_indexes）
-
-其余（Adapter 契约翻译、MLP projector + pre_norm、LlavaOnevision 继承 wire class）与
-Qwen3-VL ViT pipeline 完全一致。两条 pipeline 并存意义：DeepStack ablation 天然实验组。
-
-类层级：
-  Qwen3_5ViTBackbone(Qwen3_5VisionModel)      — 去 merger，保持 NaViT 契约
-  Qwen3_5ViTAsSiglipAdapter(nn.Module)        — 持有 Backbone，做 SigLIP ↔ NaViT 契约翻译
-
-三方对比公平性：定 384×384 AnyRes tile + 同款 projector 骨架 + 同款 Qwen3-1.7B LLM。
-"""
-
-import math
-import os
-import sys
-from typing import Optional
-
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-from transformers import (
-    AutoConfig,
-    AutoModel,
-    AutoModelForCausalLM,
-    LlavaOnevisionConfig,
-    LlavaOnevisionForConditionalGeneration,
-    LlavaOnevisionModel,
-    LlavaOnevisionPreTrainedModel,
-    Qwen3Config,
-)
-from transformers.activations import ACT2FN
-from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
-from transformers.models.qwen3_5.configuration_qwen3_5 import Qwen3_5VisionConfig
-from transformers.models.qwen3_5.modeling_qwen3_5 import Qwen3_5VisionModel
-
-# Shared layout-permutation utility lives in declip_qwenvit (single source of
-# truth — same code path runs in declip-training-side qk_cosine reorder).
-# Add VisionEncoder repo root to sys.path so this modeling file is importable
-# even when the package isn't pip-installed (ms-swift integration loads it
-# via dynamic plugin path).
-_REPO_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
-if _REPO_ROOT not in sys.path:
-    sys.path.append(_REPO_ROOT)
-from declip_qwenvit.model.qwen3vit_qk import block_merge_to_row_major_permutation  # noqa: E402
-
-
-class LlavaQwen3_5ViTConfig(LlavaOnevisionConfig):
-    """LlavaOnevisionConfig 子类，vision_config 类型换成 Qwen3_5VisionConfig。
-
-    与 Qwen3-VL ViT 版的差异：
-      - sub_configs.vision_config 用 Qwen3_5VisionConfig
-      - 不再设置 deepstack_visual_indexes（Qwen3_5VisionConfig 用 AttributeError 哨兵屏蔽此字段）
-
-    其余同 LlavaQwen3VLViTConfig（tile_size 默认 384；vision_feature_select_strategy='full'
-    必须 override，Qwen3.5 ViT 无 CLS token）。
-    """
-
-    model_type = "llava_qwen3_5vit_qwen3"
-    sub_configs = {"vision_config": Qwen3_5VisionConfig, "text_config": Qwen3Config}
-
-    def __init__(
-        self,
-        vision_config=None,
-        text_config=None,
-        tile_size: int = 384,
-        vit_register_ring: int = 0,
-        vit_skip_last_blocks: int = 0,
-        **kwargs,
-    ):
-        if isinstance(vision_config, dict):
-            vision_config = Qwen3_5VisionConfig(**vision_config)
-        elif vision_config is None:
-            vision_config = Qwen3_5VisionConfig()
-        # UniRefiner deploy-side register ring width in tokens (V10.0.1+). The S0
-        # refinement optimizes image tokens INSIDE a register-ring canvas; >0 makes
-        # every tower forward pad a zero ring around each tile/frame and drop the
-        # ring tokens after encoding (LLM-side token contract unchanged). Persisted
-        # in config.json so S2-resume and eval inherit it automatically. 0 = stock
-        # behavior (all non-V10.0.1 chains).
-        self.vit_register_ring = int(vit_register_ring)
-        # -N layer ablation (stock_minus2 chains, 2026-06-10): physically DELETE the
-        # last N ViT blocks after construction (LLaVA-NeXT convention,
-        # `del encoder.layers[-1:]`), so last_hidden_state = hidden_states[-(N+1)]
-        # of the full tower, still through final_layernorm. Physical deletion (vs
-        # forward-skip) keeps the state_dict 1:1 with truncated declip S0 ckpts
-        # (missing=0 contract) and makes silently running the dropped block
-        # impossible. Persisted in config.json; S2-resume/eval inherit. 0 = stock.
-        # NOTE: vision_config.depth stays at the full count; truncation applies at
-        # adapter build, before any load_state_dict.
-        self.vit_skip_last_blocks = int(vit_skip_last_blocks)
-        # WHY 无 `vision_config.deepstack_visual_indexes = []`（对比 Qwen3-VL ViT 版）：
-        # Qwen3_5VisionConfig 父类用 AttributeError() 哨兵显式屏蔽此字段，设置会报 AttributeError
-        # LlavaOnevision.pack_image_features 用这个作为 tile 像素大小（不是 patch_size）
-        vision_config.image_size = tile_size
-
-        if isinstance(text_config, dict):
-            text_config = Qwen3Config(**text_config)
-        elif text_config is None:
-            text_config = Qwen3Config()
-
-        # 父类默认 select_strategy='default' 会跳首 token (CLS) — Qwen3.5 ViT 无 CLS 必须用 'full'
-        kwargs.setdefault("vision_feature_select_strategy", "full")
-        super().__init__(vision_config=vision_config, text_config=text_config, **kwargs)
-
-
-class Qwen3_5ViTBackbone(Qwen3_5VisionModel):
-    """Qwen3.5 Vision 去除原生 patch_merger 的 backbone 版本（V6 final_layernorm fix (2026-05-16): append final_layernorm）。
-
-    构造时把 merger.norm 的预训练权重抠到 final_layernorm，然后 `del self.merger`
-    释放 ~37M 参数（保留 norm 的 LN 焊到末端做 post_layernorm 角色，丢弃 spatial
-    shuffle + linear_fc1/fc2，那对应 LlavaOV projector 的职责）。
-
-    架构对称（V6 final_layernorm fix (2026-05-16) 修复）：
-      SigLIP2:    encoder → post_layernorm → last_hidden_state → LlavaOV MLP → LLM
-      V6 final_layernorm fix (2026-05-16):     encoder → final_layernorm → last_hidden_state → LlavaOV MLP → LLM
-
-    forward 跑完 transformer blocks 后过 final_layernorm，再返回。下游 LlavaOnevision
-    pack_image_features 的 AnyRes 2×2 pool 接管原 merger 的空间合并职责。
-
-    输入输出契约与父类 Qwen3_5VisionModel 一致（NaViT flat）：
-      forward(hidden_states=[L, patch_dim], grid_thw=[N, 3])
-      → BaseModelOutput(last_hidden_state=[L, hidden_size])
-
-    forward 主体 1:1 对照 `Qwen3_5VisionModel.forward`（已无 deepstack loop，比
-    Qwen3VLVisionModel.forward 更短），仅跳过末尾 `self.merger(x)`，改为 final_layernorm。
-    """
-
-    def __init__(self, config):
-        super().__init__(config)
-        # V6 final_layernorm fix (2026-05-16): extract merger.norm pretrained weights into final_layernorm.
-        # Default init: even if ckpt 阶段没 inject final_layernorm.* (e.g. stock
-        # bootstrap path that filters merger.*), final_layernorm 仍持有 merger.norm
-        # 的预训练值, 不是 random — 这是防止 silent corruption 的兜底.
-        ln_w = self.merger.norm.weight.detach().clone()
-        ln_b = self.merger.norm.bias.detach().clone()
-        del self.merger
-        self.final_layernorm = nn.LayerNorm(config.hidden_size, eps=1e-6)
-        self.final_layernorm.weight.data.copy_(ln_w)
-        self.final_layernorm.bias.data.copy_(ln_b)
-
-    def forward(self, hidden_states, grid_thw, **kwargs):
-        hidden_states = self.patch_embed(hidden_states)
-
-        pos_embeds = self.fast_pos_embed_interpolate(grid_thw)
-        hidden_states = hidden_states + pos_embeds
-
-        rotary_pos_emb = self.rot_pos_emb(grid_thw)
-        seq_len, _ = hidden_states.size()
-        rotary_pos_emb = rotary_pos_emb.reshape(seq_len, -1)
-        emb = torch.cat((rotary_pos_emb, rotary_pos_emb), dim=-1)
-        position_embeddings = (emb.cos(), emb.sin())
-
-        cu_seqlens = torch.repeat_interleave(
-            grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]
-        ).cumsum(dim=0, dtype=torch.int32)
-        cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
-
-        for blk in self.blocks:
-            hidden_states = blk(
-                hidden_states,
-                cu_seqlens=cu_seqlens,
-                position_embeddings=position_embeddings,
-                **kwargs,
-            )
-
-        # V6 final_layernorm fix (2026-05-16): appended final LayerNorm — mirrors SigLIP2's post_layernorm.
-        # Per-token affine; layout-invariant (reorder happens in adapter).
-        hidden_states = self.final_layernorm(hidden_states)
-
-        return BaseModelOutput(last_hidden_state=hidden_states)
-
-
-class Qwen3_5ViTAsSiglipAdapter(nn.Module):
-    """SigLIP 契约 → NaViT 契约的翻译层。持有 Qwen3_5ViTBackbone。
-
-    对外暴露 SigLIP 式 forward(pixel_values=[N,3,H,W]) → BaseModelOutputWithPooling，
-    供 LlavaOnevision 消费；对内按官方 _preprocess 的 reshape 链把 pixel_values
-    转成 NaViT flat + grid_thw 喂给 Backbone。
-
-    reshape 链 1:1 照抄 transformers 官方 Qwen2VLImageProcessorFast._preprocess
-    (video_processing_qwen3_vl.py L227-252) —— Qwen3.5 无独立 image_processor，复用 Qwen3-VL 格式。
-    """
-
-    def __init__(self, vision_config: Qwen3_5VisionConfig, register_ring: int = 0,
-                 skip_last_blocks: int = 0):
-        super().__init__()
-        self.vision = Qwen3_5ViTBackbone(vision_config)
-        self.config = vision_config
-        # Deploy-side UniRefiner register ring (tokens). See LlavaQwen3_5ViTConfig.
-        # Ring must keep the padded grid even for spatial_merge_size=2 NaViT
-        # flattening: 24 + 2*ring stays even for any integer ring.
-        self.register_ring = int(register_ring)
-        # -N layer ablation: physically delete the last N blocks (see
-        # LlavaQwen3_5ViTConfig.vit_skip_last_blocks). final_layernorm stays,
-        # applied to the new last block's output.
-        skip = int(skip_last_blocks)
-        if skip > 0:
-            self.vision.blocks = self.vision.blocks[:-skip]
-        self.skip_last_blocks = skip
-
-    @property
-    def dtype(self):
-        return next(self.parameters()).dtype
-
-    @property
-    def device(self):
-        return next(self.parameters()).device
-
-    def _flatten_navit(self, pixel_values: torch.Tensor):
-        """[N, 3, H, W] → (flat=[N*L, patch_dim], grid_thw=[N, 3], shape=(N, L)).
-
-        L = grid_t * grid_h * grid_w = 1 * (H/16) * (W/16)
-        patch_dim = C * temporal_patch_size * patch_size^2 = 3 * 2 * 16 * 16 = 1536
-        """
-        pixel_values = pixel_values.to(dtype=self.dtype)
-        tps = self.config.temporal_patch_size
-        ps = self.config.patch_size
-        ms = self.config.spatial_merge_size
-
-        patches = pixel_values.unsqueeze(1)
-        # 对单帧图像 T=1, pad=1 → expand 一帧使 T 整除 temporal_patch_size，
-        # Conv3d 在复制帧上退化为等效 2D Conv（数学无损）
-        T = patches.shape[1]
-        pad = -T % tps
-        if pad:
-            repeats = patches[:, -1:].expand(-1, pad, -1, -1, -1)
-            patches = torch.cat((patches, repeats), dim=1)
-
-        batch_size, t, channel, H, W = patches.shape
-        grid_t = t // tps
-        grid_h = H // ps
-        grid_w = W // ps
-
-        patches = patches.view(
-            batch_size, grid_t, tps, channel,
-            grid_h // ms, ms, ps,
-            grid_w // ms, ms, ps,
-        )
-        patches = patches.permute(0, 1, 4, 7, 5, 8, 3, 2, 6, 9)
-        flatten_patches = patches.reshape(
-            batch_size,
-            grid_t * grid_h * grid_w,
-            channel * tps * ps * ps,
-        )
-
-        seq_len = grid_t * grid_h * grid_w
-        flat = flatten_patches.reshape(batch_size * seq_len, -1)
-        # on-device 构造小 tensor 再 expand，host→GPU 同步量 O(3) 而非 O(N*3)
-        grid_unit = torch.tensor(
-            [grid_t, grid_h, grid_w], dtype=torch.int32, device=pixel_values.device,
-        )
-        grid_thw = grid_unit.unsqueeze(0).expand(batch_size, -1).contiguous()
-        return flat, grid_thw, (batch_size, seq_len)
-
-    def forward(
-        self,
-        pixel_values: torch.Tensor,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-        **kwargs,
-    ) -> BaseModelOutputWithPooling:
-        ring = self.register_ring
-        if ring > 0:
-            # UniRefiner deploy protocol: surround each tile/frame with a register
-            # ring, encode, drop the ring afterwards. zero-fill in normalized pixel
-            # space == the S0 training ring (register_fill: zero, gray patches).
-            pad_px = ring * self.config.patch_size
-            pixel_values = F.pad(pixel_values, (pad_px, pad_px, pad_px, pad_px), value=0.0)
-
-        flat, grid_thw, (N, S) = self._flatten_navit(pixel_values)
-        vision_out = self.vision(flat, grid_thw=grid_thw)
-        hidden = vision_out.last_hidden_state.view(N, S, -1)
-
-        # Block-merge → row-major reorder before handing to LlavaOnevision.
-        # Internally the ViT runs in Qwen NaViT block-merge layout (pretrained
-        # pos_embed + RoPE contract); downstream LlavaOV `pack_image_features`
-        # (multi-tile AnyRes path, view(num_patch_h, num_patch_w, h, w, -1))
-        # and `apply_pooling` (video path, view(B, h, w, -1) + bilinear) BOTH
-        # assume row-major. Without this reorder, the multi-tile/video spatial
-        # pool pulls together tokens that are NOT spatially adjacent — silent
-        # corruption that doesn't fire on S1 single-tile path (line 348-351 of
-        # modeling_llava_onevision.py just flattens [N,D] verbatim) but kills
-        # S2 / eval quality.
-        grid_h = int(grid_thw[0, 1].item())
-        grid_w = int(grid_thw[0, 2].item())
-        ms = getattr(self.config, "spatial_merge_size", 2)
-        layout_perm = block_merge_to_row_major_permutation(
-            grid_h, grid_w, ms=ms, device=hidden.device,
-        )
-        hidden = hidden[:, layout_perm, :]
-
-        if ring > 0:
-            # Drop the ring tokens after the row-major reorder: keep the inner
-            # (grid_h-2r)x(grid_w-2r) block. Downstream LlavaOV consumes the same
-            # row-major image-token count as the ring-off path.
-            keep = torch.ones(grid_h, grid_w, dtype=torch.bool, device=hidden.device)
-            keep[:ring, :] = False
-            keep[grid_h - ring:, :] = False
-            keep[:, :ring] = False
-            keep[:, grid_w - ring:] = False
-            hidden = hidden[:, keep.reshape(-1), :]
-
-        return BaseModelOutputWithPooling(
-            last_hidden_state=hidden,
-            # LlavaOnevision 索引 hidden_states[vision_feature_layer=-1]；tuple 长度 1 足够
-            hidden_states=(hidden,),
-            pooler_output=None,
-        )
-
-
-class LlavaQwen3_5ViTMultiModalProjector(nn.Module):
-    """标准 LlavaOnevision projector（V6 final_layernorm fix (2026-05-16): pre_norm → Identity）。
-
-    V6 final_layernorm fix (2026-05-16) 修复后, encoder 末端已自带 final_layernorm（与 SigLIP2 post_layernorm 对称），
-    projector 不再需要补 LN — pre_norm 改为 nn.Identity，对齐 SigLIP2 plugin 的
-    LlavaOnevision stock projector 结构（裸 linear_1 → GELU → linear_2），
-    保证 SigLIP2 / Qwen3.5 / Qwen3-VL 三个 backbone 在 LlavaOV 设定下公平对比。
-
-    （历史：V6.0.0~V6.0.4 时期 encoder 无 final LN，projector pre_norm 是补丁；
-    现在补丁回到 encoder 内部，projector 回归 stock 形态。）
-    """
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        super().__init__()
-        num_feature_layers = (
-            1 if isinstance(config.vision_feature_layer, int) else len(config.vision_feature_layer)
-        )
-        vision_dim = config.vision_config.hidden_size * num_feature_layers
-        text_dim = config.text_config.hidden_size
-        bias = getattr(config, "multimodal_projector_bias", True)
-
-        # V6 final_layernorm fix (2026-05-16): pre_norm = Identity (encoder 已自带 final_layernorm).
-        self.pre_norm = nn.Identity()
-        self.linear_1 = nn.Linear(vision_dim, text_dim, bias=bias)
-        self.act = ACT2FN[config.projector_hidden_act]
-        self.linear_2 = nn.Linear(text_dim, text_dim, bias=bias)
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        return self.linear_2(self.act(self.linear_1(self.pre_norm(x))))
-
-
-class LlavaQwen3_5ViTModel(LlavaOnevisionModel):
-    """继承 LlavaOnevisionModel 但绕过其 __init__ 手动装配。
-
-    父类 __init__ 调 `AutoModel.from_config(config.vision_config)` 会对 Qwen3_5VisionConfig
-    抛 "Unrecognized configuration"（Qwen3.5 vision 没注册到 AutoModel）。手动装配避开
-    这一步，同时省掉"先构造 Qwen3_5VisionModel 再被替换"的双重开销（~1.3GB init-peak）。
-
-    装配顺序与父类一致：vision_tower / projector / image_newline / language_model / post_init。
-    """
-
-    config_class = LlavaQwen3_5ViTConfig
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        # 跳过 LlavaOnevisionModel.__init__（AutoModel 不识别 Qwen3_5VisionConfig）
-        LlavaOnevisionPreTrainedModel.__init__(self, config)
-        self.vision_tower = Qwen3_5ViTAsSiglipAdapter(
-            config.vision_config,
-            register_ring=getattr(config, "vit_register_ring", 0),
-            skip_last_blocks=getattr(config, "vit_skip_last_blocks", 0),
-        )
-        self.multi_modal_projector = LlavaQwen3_5ViTMultiModalProjector(config)
-        embed_std = 1 / math.sqrt(config.text_config.hidden_size)
-        self.image_newline = nn.Parameter(
-            torch.randn(config.text_config.hidden_size, dtype=self.dtype) * embed_std
-        )
-        self.vocab_size = config.text_config.vocab_size
-        self.language_model = AutoModel.from_config(config.text_config)
-        self.post_init()
-
-
-class LlavaQwen3_5ViTForConditionalGeneration(LlavaOnevisionForConditionalGeneration):
-    """继承 LlavaOnevisionForConditionalGeneration，只换 self.model。
-
-    同样跳过父类 __init__（避免重复构造 LlavaOnevisionModel，根因见 LlavaQwen3_5ViTModel）。
-    """
-
-    config_class = LlavaQwen3_5ViTConfig
-
-    def __init__(self, config: LlavaQwen3_5ViTConfig):
-        LlavaOnevisionPreTrainedModel.__init__(self, config)
-        self.model = LlavaQwen3_5ViTModel(config)
-        self.lm_head = nn.Linear(
-            config.text_config.hidden_size, config.text_config.vocab_size, bias=False
-        )
-        self.post_init()
-
-
-AutoConfig.register(LlavaQwen3_5ViTConfig.model_type, LlavaQwen3_5ViTConfig)
-AutoModelForCausalLM.register(LlavaQwen3_5ViTConfig, LlavaQwen3_5ViTForConditionalGeneration)
-
-
-__all__ = [
-    "LlavaQwen3_5ViTConfig",
-    "Qwen3_5ViTBackbone",
-    "Qwen3_5ViTAsSiglipAdapter",
-    "LlavaQwen3_5ViTMultiModalProjector",
-    "LlavaQwen3_5ViTModel",
-    "LlavaQwen3_5ViTForConditionalGeneration",
-]
diff --git a/ckpts/v10_2_32f/processor_config.json b/ckpts/v10_2_32f/processor_config.json
deleted file mode 100644
index 204ed5d216c648cbd71da6c25d67b4bf1ec84433..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/processor_config.json
+++ /dev/null
@@ -1,203 +0,0 @@
-{
-  "image_processor": {
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_pad": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "image_grid_pinpoints": [
-      [
-        384,
-        384
-      ],
-      [
-        384,
-        768
-      ],
-      [
-        384,
-        1152
-      ],
-      [
-        384,
-        1536
-      ],
-      [
-        384,
-        1920
-      ],
-      [
-        384,
-        2304
-      ],
-      [
-        768,
-        384
-      ],
-      [
-        768,
-        768
-      ],
-      [
-        768,
-        1152
-      ],
-      [
-        768,
-        1536
-      ],
-      [
-        768,
-        1920
-      ],
-      [
-        768,
-        2304
-      ],
-      [
-        1152,
-        384
-      ],
-      [
-        1152,
-        768
-      ],
-      [
-        1152,
-        1152
-      ],
-      [
-        1152,
-        1536
-      ],
-      [
-        1152,
-        1920
-      ],
-      [
-        1152,
-        2304
-      ],
-      [
-        1536,
-        384
-      ],
-      [
-        1536,
-        768
-      ],
-      [
-        1536,
-        1152
-      ],
-      [
-        1536,
-        1536
-      ],
-      [
-        1536,
-        1920
-      ],
-      [
-        1536,
-        2304
-      ],
-      [
-        1920,
-        384
-      ],
-      [
-        1920,
-        768
-      ],
-      [
-        1920,
-        1152
-      ],
-      [
-        1920,
-        1536
-      ],
-      [
-        1920,
-        1920
-      ],
-      [
-        1920,
-        2304
-      ],
-      [
-        2304,
-        384
-      ],
-      [
-        2304,
-        768
-      ],
-      [
-        2304,
-        1152
-      ],
-      [
-        2304,
-        1536
-      ],
-      [
-        2304,
-        1920
-      ],
-      [
-        2304,
-        2304
-      ]
-    ],
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_processor_type": "LlavaOnevisionImageProcessor",
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "size": {
-      "height": 384,
-      "width": 384
-    }
-  },
-  "image_token": "<image>",
-  "num_image_tokens": 576,
-  "processor_class": "LlavaOnevisionProcessor",
-  "video_processor": {
-    "do_convert_rgb": true,
-    "do_normalize": true,
-    "do_rescale": true,
-    "do_resize": true,
-    "do_sample_frames": false,
-    "image_mean": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "image_std": [
-      0.5,
-      0.5,
-      0.5
-    ],
-    "resample": 3,
-    "rescale_factor": 0.00392156862745098,
-    "return_metadata": false,
-    "size": {
-      "height": 384,
-      "width": 384
-    },
-    "video_processor_type": "LlavaOnevisionVideoProcessor"
-  },
-  "video_token": "<video>",
-  "vision_aspect_ratio": "anyres_max_9",
-  "vision_feature_select_strategy": null
-}
diff --git a/ckpts/v10_2_32f/tokenizer.json b/ckpts/v10_2_32f/tokenizer.json
deleted file mode 100644
index c40d89c9226d696c215d65a68eea68c6dd55d79e..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/tokenizer.json
+++ /dev/null
@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:9024318c850eaadf26be79389d21b07a7afd8f1af749b89f9109b06c0d12173c
-size 11423018
diff --git a/ckpts/v10_2_32f/tokenizer_config.json b/ckpts/v10_2_32f/tokenizer_config.json
deleted file mode 100644
index 91a90671a687fb5d1fc3ed576d216e5e67d9e11b..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/tokenizer_config.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "add_prefix_space": false,
-  "backend": "tokenizers",
-  "bos_token": null,
-  "clean_up_tokenization_spaces": false,
-  "eos_token": "<|im_end|>",
-  "errors": "replace",
-  "extra_special_tokens": [
-    "<image>",
-    "<video>"
-  ],
-  "is_local": true,
-  "model_max_length": 1010000,
-  "pad_token": "<|endoftext|>",
-  "processor_class": "LlavaOnevisionProcessor",
-  "split_special_tokens": false,
-  "tokenizer_class": "Qwen2Tokenizer",
-  "unk_token": null
-}
diff --git a/ckpts/v10_2_32f/zero_to_fp32.py b/ckpts/v10_2_32f/zero_to_fp32.py
deleted file mode 100644
index 5995d6e6f04e43b989587aa9022a3aef0c66d694..0000000000000000000000000000000000000000
--- a/ckpts/v10_2_32f/zero_to_fp32.py
+++ /dev/null
@@ -1,760 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright (c) Microsoft Corporation.
-# SPDX-License-Identifier: Apache-2.0
-
-# DeepSpeed Team
-
-# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
-# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
-# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
-# application.
-#
-# example:
-#   python zero_to_fp32.py . output_dir/
-#   or
-#   python zero_to_fp32.py . output_dir/ --safe_serialization
-
-import argparse
-import torch
-import glob
-import math
-import os
-import re
-import gc
-import json
-import numpy as np
-from tqdm import tqdm
-from collections import OrderedDict
-from dataclasses import dataclass
-
-# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
-# DeepSpeed data structures it has to be available in the current python environment.
-from deepspeed.utils import logger
-from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
-                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
-                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
-
-
-@dataclass
-class zero_model_state:
-    buffers: dict()
-    param_shapes: dict()
-    shared_params: list
-    ds_version: int
-    frozen_param_shapes: dict()
-    frozen_param_fragments: dict()
-
-
-debug = 0
-
-# load to cpu
-device = torch.device('cpu')
-
-
-def atoi(text):
-    return int(text) if text.isdigit() else text
-
-
-def natural_keys(text):
-    '''
-    alist.sort(key=natural_keys) sorts in human order
-    http://nedbatchelder.com/blog/200712/human_sorting.html
-    (See Toothy's implementation in the comments)
-    '''
-    return [atoi(c) for c in re.split(r'(\d+)', text)]
-
-
-def get_model_state_file(checkpoint_dir, zero_stage):
-    if not os.path.isdir(checkpoint_dir):
-        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
-
-    # there should be only one file
-    if zero_stage <= 2:
-        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
-    elif zero_stage == 3:
-        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
-
-    if not os.path.exists(file):
-        raise FileNotFoundError(f"can't find model states file at '{file}'")
-
-    return file
-
-
-def get_checkpoint_files(checkpoint_dir, glob_pattern):
-    # XXX: need to test that this simple glob rule works for multi-node setup too
-    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
-
-    if len(ckpt_files) == 0:
-        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
-
-    return ckpt_files
-
-
-def get_optim_files(checkpoint_dir):
-    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
-
-
-def get_model_state_files(checkpoint_dir):
-    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
-
-
-def parse_model_states(files):
-    zero_model_states = []
-    for file in files:
-        state_dict = torch.load(file, map_location=device, weights_only=False)
-
-        if BUFFER_NAMES not in state_dict:
-            raise ValueError(f"{file} is not a model state checkpoint")
-        buffer_names = state_dict[BUFFER_NAMES]
-        if debug:
-            print("Found buffers:", buffer_names)
-
-        # recover just the buffers while restoring them to fp32 if they were saved in fp16
-        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
-        param_shapes = state_dict[PARAM_SHAPES]
-
-        # collect parameters that are included in param_shapes
-        param_names = []
-        for s in param_shapes:
-            for name in s.keys():
-                param_names.append(name)
-
-        # update with frozen parameters
-        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
-        if frozen_param_shapes is not None:
-            if debug:
-                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
-            param_names += list(frozen_param_shapes.keys())
-
-        # handle shared params
-        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
-
-        ds_version = state_dict.get(DS_VERSION, None)
-
-        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
-
-        z_model_state = zero_model_state(buffers=buffers,
-                                         param_shapes=param_shapes,
-                                         shared_params=shared_params,
-                                         ds_version=ds_version,
-                                         frozen_param_shapes=frozen_param_shapes,
-                                         frozen_param_fragments=frozen_param_fragments)
-        zero_model_states.append(z_model_state)
-
-    return zero_model_states
-
-
-def parse_optim_states(files, ds_checkpoint_dir):
-    total_files = len(files)
-    state_dicts = []
-    for f in tqdm(files, desc='Loading checkpoint shards'):
-        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
-        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
-        # and also handle the case where it was already removed by another helper script
-        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
-        state_dicts.append(state_dict)
-
-    if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
-        raise ValueError(f"{files[0]} is not a zero checkpoint")
-    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
-    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
-
-    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
-    # parameters can be different from data parallelism for non-expert parameters. So we can just
-    # use the max of the partition_count to get the dp world_size.
-
-    if type(world_size) is list:
-        world_size = max(world_size)
-
-    if world_size != total_files:
-        raise ValueError(
-            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
-            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
-        )
-
-    # the groups are named differently in each stage
-    if zero_stage <= 2:
-        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
-    elif zero_stage == 3:
-        fp32_groups_key = FP32_FLAT_GROUPS
-    else:
-        raise ValueError(f"unknown zero stage {zero_stage}")
-
-    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
-    return zero_stage, world_size, fp32_flat_groups
-
-
-def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
-    """
-    Returns fp32 state_dict reconstructed from ds checkpoint
-
-    Args:
-        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
-
-    """
-    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
-
-    optim_files = get_optim_files(ds_checkpoint_dir)
-    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
-    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
-
-    model_files = get_model_state_files(ds_checkpoint_dir)
-
-    zero_model_states = parse_model_states(model_files)
-    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
-
-    if zero_stage <= 2:
-        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                                          exclude_frozen_parameters)
-    elif zero_stage == 3:
-        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                                          exclude_frozen_parameters)
-
-
-def _zero2_merge_frozen_params(state_dict, zero_model_states):
-    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
-        return
-
-    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
-    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
-
-    if debug:
-        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
-        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
-
-        wanted_params = len(frozen_param_shapes)
-        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
-        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
-        print(f'Frozen params: Have {avail_numel} numels to process.')
-        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
-
-    total_params = 0
-    total_numel = 0
-    for name, shape in frozen_param_shapes.items():
-        total_params += 1
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-
-        state_dict[name] = frozen_param_fragments[name]
-
-        if debug:
-            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
-
-    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _has_callable(obj, fn):
-    attr = getattr(obj, fn, None)
-    return callable(attr)
-
-
-def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
-    param_shapes = zero_model_states[0].param_shapes
-
-    # Reconstruction protocol:
-    #
-    # XXX: document this
-
-    if debug:
-        for i in range(world_size):
-            for j in range(len(fp32_flat_groups[0])):
-                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
-
-    # XXX: memory usage doubles here (zero2)
-    num_param_groups = len(fp32_flat_groups[0])
-    merged_single_partition_of_fp32_groups = []
-    for i in range(num_param_groups):
-        merged_partitions = [sd[i] for sd in fp32_flat_groups]
-        full_single_fp32_vector = torch.cat(merged_partitions, 0)
-        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
-    avail_numel = sum(
-        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
-
-    if debug:
-        wanted_params = sum([len(shapes) for shapes in param_shapes])
-        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
-        # not asserting if there is a mismatch due to possible padding
-        print(f"Have {avail_numel} numels to process.")
-        print(f"Need {wanted_numel} numels in {wanted_params} params.")
-
-    # params
-    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
-    # out-of-core computing solution
-    total_numel = 0
-    total_params = 0
-    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
-        offset = 0
-        avail_numel = full_single_fp32_vector.numel()
-        for name, shape in shapes.items():
-
-            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
-            total_numel += unpartitioned_numel
-            total_params += 1
-
-            if debug:
-                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
-            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
-            offset += unpartitioned_numel
-
-        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
-        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
-        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
-        # live optimizer object, so we are checking that the numbers are within the right range
-        align_to = 2 * world_size
-
-        def zero2_align(x):
-            return align_to * math.ceil(x / align_to)
-
-        if debug:
-            print(f"original offset={offset}, avail_numel={avail_numel}")
-
-        offset = zero2_align(offset)
-        avail_numel = zero2_align(avail_numel)
-
-        if debug:
-            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
-
-        # Sanity check
-        if offset != avail_numel:
-            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
-
-    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                               exclude_frozen_parameters):
-    state_dict = OrderedDict()
-
-    # buffers
-    buffers = zero_model_states[0].buffers
-    state_dict.update(buffers)
-    if debug:
-        print(f"added {len(buffers)} buffers")
-
-    if not exclude_frozen_parameters:
-        _zero2_merge_frozen_params(state_dict, zero_model_states)
-
-    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
-
-    # recover shared parameters
-    for pair in zero_model_states[0].shared_params:
-        if pair[1] in state_dict:
-            state_dict[pair[0]] = state_dict[pair[1]]
-
-    return state_dict
-
-
-def zero3_partitioned_param_info(unpartitioned_numel, world_size):
-    remainder = unpartitioned_numel % world_size
-    padding_numel = (world_size - remainder) if remainder else 0
-    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
-    return partitioned_numel, padding_numel
-
-
-def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
-    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
-        return
-
-    if debug:
-        for i in range(world_size):
-            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
-            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
-
-        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
-        wanted_params = len(frozen_param_shapes)
-        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
-        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
-        print(f'Frozen params: Have {avail_numel} numels to process.')
-        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
-
-    total_params = 0
-    total_numel = 0
-    for name, shape in zero_model_states[0].frozen_param_shapes.items():
-        total_params += 1
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-
-        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
-        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
-
-        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
-
-        if debug:
-            print(
-                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
-            )
-
-    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
-
-
-class GatheredTensor:
-    """
-    A pseudo tensor that collects partitioned weights.
-    It is more memory efficient when there are multiple groups.
-    """
-
-    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
-        self.flat_groups = flat_groups
-        self.flat_groups_offset = flat_groups_offset
-        self.offset = offset
-        self.partitioned_numel = partitioned_numel
-        self.shape = shape
-        self.dtype = self.flat_groups[0][0].dtype
-
-    def contiguous(self):
-        """
-        Merge partitioned weights from flat_groups into a single tensor.
-        """
-        end_idx = self.offset + self.partitioned_numel
-        world_size = len(self.flat_groups)
-        pad_flat_param_chunks = []
-
-        for rank_i in range(world_size):
-            # for each rank, we need to collect weights from related group/groups
-            flat_groups_at_rank_i = self.flat_groups[rank_i]
-            start_group_id = None
-            end_group_id = None
-            for group_id in range(len(self.flat_groups_offset)):
-                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
-                    start_group_id = group_id
-                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
-                    end_group_id = group_id
-                    break
-            # collect weights from related group/groups
-            for group_id in range(start_group_id, end_group_id + 1):
-                flat_tensor = flat_groups_at_rank_i[group_id]
-                start_offset = self.offset - self.flat_groups_offset[group_id]
-                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
-                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
-
-        # collect weights from all ranks
-        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
-        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
-        return param
-
-
-def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
-    param_shapes = zero_model_states[0].param_shapes
-    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
-
-    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
-    # param, re-consolidating each param, while dealing with padding if any
-
-    # merge list of dicts, preserving order
-    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
-
-    if debug:
-        for i in range(world_size):
-            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
-
-        wanted_params = len(param_shapes)
-        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
-        # not asserting if there is a mismatch due to possible padding
-        avail_numel = fp32_flat_groups[0].numel() * world_size
-        print(f"Trainable params: Have {avail_numel} numels to process.")
-        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
-
-    # params
-    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
-    # out-of-core computing solution
-    offset = 0
-    total_numel = 0
-    total_params = 0
-    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
-    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
-        unpartitioned_numel = shape.numel()
-        total_numel += unpartitioned_numel
-        total_params += 1
-        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
-
-        if debug:
-            print(
-                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
-            )
-
-        # memory efficient tensor
-        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
-        state_dict[name] = tensor
-        offset += partitioned_numel
-
-    offset *= world_size
-
-    # Sanity check
-    if offset != avail_numel:
-        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
-
-    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
-
-
-def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
-                                               exclude_frozen_parameters):
-    state_dict = OrderedDict()
-
-    # buffers
-    buffers = zero_model_states[0].buffers
-    state_dict.update(buffers)
-    if debug:
-        print(f"added {len(buffers)} buffers")
-
-    if not exclude_frozen_parameters:
-        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
-
-    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
-
-    # recover shared parameters
-    for pair in zero_model_states[0].shared_params:
-        if pair[1] in state_dict:
-            state_dict[pair[0]] = state_dict[pair[1]]
-
-    return state_dict
-
-
-def to_torch_tensor(state_dict, return_empty_tensor=False):
-    """
-    Convert state_dict of GatheredTensor to torch tensor
-    """
-    torch_state_dict = {}
-    converted_tensors = {}
-    for name, tensor in state_dict.items():
-        tensor_id = id(tensor)
-        if tensor_id in converted_tensors:  # shared tensors
-            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
-            torch_state_dict[name] = shared_tensor
-        else:
-            converted_tensors[tensor_id] = name
-            if return_empty_tensor:
-                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
-            else:
-                torch_state_dict[name] = tensor.contiguous()
-    return torch_state_dict
-
-
-def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
-                                             tag=None,
-                                             exclude_frozen_parameters=False,
-                                             lazy_mode=False):
-    """
-    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
-    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
-    via a model hub.
-
-    Args:
-        - ``checkpoint_dir``: path to the desired checkpoint folder
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
-        - ``exclude_frozen_parameters``: exclude frozen parameters
-        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
-          Convert the pesduo tensor to torch tensor by ``.contiguous()``
-
-    Returns:
-        - pytorch ``state_dict``
-
-    A typical usage might be ::
-
-        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
-        # do the training and checkpoint saving
-        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
-        model = model.cpu() # move to cpu
-        model.load_state_dict(state_dict)
-        # submit to model hub or save the model to share with others
-
-    In this example the ``model`` will no longer be usable in the deepspeed context of the same
-    application. i.e. you will need to re-initialize the deepspeed engine, since
-    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
-
-    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
-
-    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
-    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
-    the checkpoint. Or you can load state_dict in lazy mode ::
-
-        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
-        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
-        for name, lazy_tensor in state_dict.item():
-            tensor = lazy_tensor.contiguous()  # to cpu
-            print(name, tensor)
-            # del tensor to release memory if it no longer in use
-    """
-    if tag is None:
-        latest_path = os.path.join(checkpoint_dir, 'latest')
-        if os.path.isfile(latest_path):
-            with open(latest_path, 'r') as fd:
-                tag = fd.read().strip()
-        else:
-            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
-
-    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
-
-    if not os.path.isdir(ds_checkpoint_dir):
-        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
-
-    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
-    if lazy_mode:
-        return state_dict
-    else:
-        return to_torch_tensor(state_dict)
-
-
-def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
-                                               output_dir,
-                                               max_shard_size="5GB",
-                                               safe_serialization=False,
-                                               tag=None,
-                                               exclude_frozen_parameters=False):
-    """
-    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
-    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
-
-    Args:
-        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
-        - ``output_dir``: directory to the pytorch fp32 state_dict output files
-        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
-        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
-        - ``exclude_frozen_parameters``: exclude frozen parameters
-    """
-
-    # Dependency pre-check
-    if safe_serialization:
-        try:
-            from safetensors.torch import save_file
-        except ImportError:
-            print('If you want to use `safe_serialization`, please `pip install safetensors`')
-            raise
-    if max_shard_size is not None:
-        try:
-            from huggingface_hub import split_torch_state_dict_into_shards
-        except ImportError:
-            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
-            raise
-
-    # Convert zero checkpoint to state_dict
-    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
-                                                          tag,
-                                                          exclude_frozen_parameters,
-                                                          lazy_mode=True)
-
-    # Shard the model if it is too big.
-    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
-    if max_shard_size is not None:
-        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
-        # an memory-efficient approach for sharding
-        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
-        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
-                                                              filename_pattern=filename_pattern,
-                                                              max_shard_size=max_shard_size)
-    else:
-        from collections import namedtuple
-        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
-        state_dict_split = StateDictSplit(is_sharded=False,
-                                          filename_to_tensors={weights_name: list(state_dict.keys())})
-
-    # Save the model by shard
-    os.makedirs(output_dir, exist_ok=True)
-    filename_to_tensors = state_dict_split.filename_to_tensors.items()
-    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
-        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
-        shard_state_dict = to_torch_tensor(shard_state_dict)
-        output_path = os.path.join(output_dir, shard_file)
-        if safe_serialization:
-            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
-        else:
-            torch.save(shard_state_dict, output_path)
-        # release the memory of current shard
-        for tensor_name in list(shard_state_dict.keys()):
-            del state_dict[tensor_name]
-            del shard_state_dict[tensor_name]
-        del shard_state_dict
-        gc.collect()
-
-    # Save index if sharded
-    if state_dict_split.is_sharded:
-        index = {
-            "metadata": state_dict_split.metadata,
-            "weight_map": state_dict_split.tensor_to_filename,
-        }
-        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
-        save_index_file = os.path.join(output_dir, save_index_file)
-        with open(save_index_file, "w", encoding="utf-8") as f:
-            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
-            f.write(content)
-
-
-def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
-    """
-    1. Put the provided model to cpu
-    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
-    3. Load it into the provided model
-
-    Args:
-        - ``model``: the model object to update
-        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
-        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
-
-    Returns:
-        - ``model`: modified model
-
-    Make sure you have plenty of CPU memory available before you call this function. If you don't
-    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
-    conveniently placed for you in the checkpoint folder.
-
-    A typical usage might be ::
-
-        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
-        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
-        # submit to model hub or save the model to share with others
-
-    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
-    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
-    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
-
-    """
-    logger.info("Extracting fp32 weights")
-    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
-
-    logger.info("Overwriting model with fp32 weights")
-    model = model.cpu()
-    model.load_state_dict(state_dict, strict=False)
-
-    return model
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument("checkpoint_dir",
-                        type=str,
-                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
-    parser.add_argument("output_dir",
-                        type=str,
-                        help="directory to the pytorch fp32 state_dict output files"
-                        "(e.g. path/checkpoint-12-output/)")
-    parser.add_argument(
-        "--max_shard_size",
-        type=str,
-        default="5GB",
-        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
-        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
-        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
-        "without CPU OOM issues.")
-    parser.add_argument(
-        "--safe_serialization",
-        default=False,
-        action='store_true',
-        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
-    parser.add_argument("-t",
-                        "--tag",
-                        type=str,
-                        default=None,
-                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
-    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
-    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
-    args = parser.parse_args()
-
-    debug = args.debug
-
-    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
-                                               args.output_dir,
-                                               max_shard_size=args.max_shard_size,
-                                               safe_serialization=args.safe_serialization,
-                                               tag=args.tag,
-                                               exclude_frozen_parameters=args.exclude_frozen_parameters)