Reorg: move 16f stock/v9_1 ckpts into S2 stage-first tree (drop machine name, tag 16f)
Browse files- .gitattributes +2 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/args.json +376 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/chat_template.jinja +61 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/chat_template.json +3 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/config.json +262 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/generation_config.json +12 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/latest +1 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/model-00001-of-00002.safetensors +3 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/model-00002-of-00002.safetensors +3 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/model.safetensors.index.json +705 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/modeling_qwen3_5vit_qwen3.py +351 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/processor_config.json +203 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/tokenizer.json +3 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/tokenizer_config.json +19 -0
- ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/zero_to_fp32.py +760 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/args.json +376 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/chat_template.jinja +61 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/chat_template.json +3 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/config.json +262 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/generation_config.json +12 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/latest +1 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/model-00001-of-00002.safetensors +3 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/model-00002-of-00002.safetensors +3 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/model.safetensors.index.json +705 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/modeling_qwen3_5vit_qwen3.py +351 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/processor_config.json +203 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/tokenizer.json +3 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/tokenizer_config.json +19 -0
- ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/zero_to_fp32.py +760 -0
.gitattributes
CHANGED
|
@@ -81,3 +81,5 @@ ckpts/S2/4b/qwen3_5_2b/v10_2_32f_10pct/v0-20260610-154054/checkpoint-785/tokeniz
|
|
| 81 |
ckpts/S2/4b/qwen3_5_2b/stock_32f_10pct/v0-20260606-044138/checkpoint-785/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 82 |
ckpts/v10_2_32f/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 83 |
ckpts/stock_32f/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
| 81 |
ckpts/S2/4b/qwen3_5_2b/stock_32f_10pct/v0-20260606-044138/checkpoint-785/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 82 |
ckpts/v10_2_32f/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 83 |
ckpts/stock_32f/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 84 |
+
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 85 |
+
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/args.json
ADDED
|
@@ -0,0 +1,376 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"output_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_qwen3_5vit_stock_baseline_a800_10pct/v0-20260530-001435",
|
| 3 |
+
"per_device_train_batch_size": 1,
|
| 4 |
+
"num_train_epochs": 3.0,
|
| 5 |
+
"max_steps": 505,
|
| 6 |
+
"learning_rate": 1e-05,
|
| 7 |
+
"lr_scheduler_type": "cosine",
|
| 8 |
+
"lr_scheduler_kwargs": null,
|
| 9 |
+
"warmup_steps": 0,
|
| 10 |
+
"optim": "adamw_torch_fused",
|
| 11 |
+
"optim_args": null,
|
| 12 |
+
"weight_decay": 0.1,
|
| 13 |
+
"adam_beta1": 0.9,
|
| 14 |
+
"adam_beta2": 0.95,
|
| 15 |
+
"adam_epsilon": 1e-08,
|
| 16 |
+
"optim_target_modules": null,
|
| 17 |
+
"gradient_accumulation_steps": 8,
|
| 18 |
+
"average_tokens_across_devices": true,
|
| 19 |
+
"max_grad_norm": 1.0,
|
| 20 |
+
"label_smoothing_factor": 0.0,
|
| 21 |
+
"bf16": true,
|
| 22 |
+
"fp16": false,
|
| 23 |
+
"bf16_full_eval": false,
|
| 24 |
+
"fp16_full_eval": false,
|
| 25 |
+
"tf32": null,
|
| 26 |
+
"gradient_checkpointing": true,
|
| 27 |
+
"gradient_checkpointing_kwargs": "{\"use_reentrant\": false}",
|
| 28 |
+
"torch_compile": false,
|
| 29 |
+
"torch_compile_backend": null,
|
| 30 |
+
"torch_compile_mode": null,
|
| 31 |
+
"use_liger_kernel": false,
|
| 32 |
+
"liger_kernel_config": null,
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"neftune_noise_alpha": null,
|
| 35 |
+
"torch_empty_cache_steps": null,
|
| 36 |
+
"auto_find_batch_size": false,
|
| 37 |
+
"logging_strategy": "steps",
|
| 38 |
+
"logging_steps": 1,
|
| 39 |
+
"logging_first_step": true,
|
| 40 |
+
"log_on_each_node": true,
|
| 41 |
+
"logging_nan_inf_filter": true,
|
| 42 |
+
"include_num_input_tokens_seen": false,
|
| 43 |
+
"log_level": "passive",
|
| 44 |
+
"log_level_replica": "warning",
|
| 45 |
+
"disable_tqdm": null,
|
| 46 |
+
"report_to": [
|
| 47 |
+
"none"
|
| 48 |
+
],
|
| 49 |
+
"run_name": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_qwen3_5vit_stock_baseline_a800_10pct/v0-20260530-001435",
|
| 50 |
+
"project": "huggingface",
|
| 51 |
+
"trackio_space_id": "trackio",
|
| 52 |
+
"eval_strategy": "no",
|
| 53 |
+
"eval_steps": 505.0,
|
| 54 |
+
"eval_delay": 0,
|
| 55 |
+
"per_device_eval_batch_size": 1,
|
| 56 |
+
"prediction_loss_only": false,
|
| 57 |
+
"eval_on_start": false,
|
| 58 |
+
"eval_do_concat_batches": true,
|
| 59 |
+
"eval_use_gather_object": false,
|
| 60 |
+
"eval_accumulation_steps": null,
|
| 61 |
+
"include_for_metrics": [],
|
| 62 |
+
"batch_eval_metrics": false,
|
| 63 |
+
"save_only_model": false,
|
| 64 |
+
"save_strategy": "steps",
|
| 65 |
+
"save_steps": 505.0,
|
| 66 |
+
"save_on_each_node": false,
|
| 67 |
+
"save_total_limit": 1,
|
| 68 |
+
"enable_jit_checkpoint": false,
|
| 69 |
+
"push_to_hub": false,
|
| 70 |
+
"hub_token": null,
|
| 71 |
+
"hub_private_repo": null,
|
| 72 |
+
"hub_model_id": null,
|
| 73 |
+
"hub_strategy": "every_save",
|
| 74 |
+
"hub_always_push": false,
|
| 75 |
+
"hub_revision": null,
|
| 76 |
+
"load_best_model_at_end": false,
|
| 77 |
+
"metric_for_best_model": "loss",
|
| 78 |
+
"greater_is_better": false,
|
| 79 |
+
"ignore_data_skip": false,
|
| 80 |
+
"restore_callback_states_from_checkpoint": false,
|
| 81 |
+
"full_determinism": false,
|
| 82 |
+
"seed": 42,
|
| 83 |
+
"data_seed": 42,
|
| 84 |
+
"use_cpu": false,
|
| 85 |
+
"accelerator_config": "{\"dispatch_batches\": false}",
|
| 86 |
+
"parallelism_config": null,
|
| 87 |
+
"dataloader_drop_last": false,
|
| 88 |
+
"dataloader_num_workers": 6,
|
| 89 |
+
"dataloader_pin_memory": true,
|
| 90 |
+
"dataloader_persistent_workers": true,
|
| 91 |
+
"dataloader_prefetch_factor": 4,
|
| 92 |
+
"remove_unused_columns": true,
|
| 93 |
+
"label_names": null,
|
| 94 |
+
"train_sampling_strategy": "random",
|
| 95 |
+
"length_column_name": "length",
|
| 96 |
+
"ddp_find_unused_parameters": null,
|
| 97 |
+
"ddp_bucket_cap_mb": null,
|
| 98 |
+
"ddp_broadcast_buffers": null,
|
| 99 |
+
"ddp_backend": null,
|
| 100 |
+
"ddp_timeout": 7200,
|
| 101 |
+
"fsdp": [],
|
| 102 |
+
"fsdp_config": null,
|
| 103 |
+
"deepspeed": {
|
| 104 |
+
"fp16": {
|
| 105 |
+
"enabled": "auto",
|
| 106 |
+
"loss_scale": 0,
|
| 107 |
+
"loss_scale_window": 1000,
|
| 108 |
+
"initial_scale_power": 16,
|
| 109 |
+
"hysteresis": 2,
|
| 110 |
+
"min_loss_scale": 1
|
| 111 |
+
},
|
| 112 |
+
"bf16": {
|
| 113 |
+
"enabled": "auto"
|
| 114 |
+
},
|
| 115 |
+
"zero_optimization": {
|
| 116 |
+
"stage": 2,
|
| 117 |
+
"offload_optimizer": {
|
| 118 |
+
"device": "none",
|
| 119 |
+
"pin_memory": true
|
| 120 |
+
},
|
| 121 |
+
"allgather_partitions": true,
|
| 122 |
+
"allgather_bucket_size": 200000000.0,
|
| 123 |
+
"overlap_comm": false,
|
| 124 |
+
"reduce_scatter": true,
|
| 125 |
+
"reduce_bucket_size": 200000000.0,
|
| 126 |
+
"contiguous_gradients": true
|
| 127 |
+
},
|
| 128 |
+
"gradient_accumulation_steps": "auto",
|
| 129 |
+
"gradient_clipping": "auto",
|
| 130 |
+
"steps_per_print": 2000,
|
| 131 |
+
"train_batch_size": "auto",
|
| 132 |
+
"train_micro_batch_size_per_gpu": "auto",
|
| 133 |
+
"wall_clock_breakdown": false
|
| 134 |
+
},
|
| 135 |
+
"debug": null,
|
| 136 |
+
"skip_memory_metrics": true,
|
| 137 |
+
"do_train": false,
|
| 138 |
+
"do_eval": false,
|
| 139 |
+
"do_predict": false,
|
| 140 |
+
"resume_from_checkpoint": null,
|
| 141 |
+
"warmup_ratio": 0.05,
|
| 142 |
+
"logging_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_qwen3_5vit_stock_baseline_a800_10pct/v0-20260530-001435/runs",
|
| 143 |
+
"local_rank": 0,
|
| 144 |
+
"sortish_sampler": false,
|
| 145 |
+
"predict_with_generate": false,
|
| 146 |
+
"generation_max_length": null,
|
| 147 |
+
"generation_num_beams": null,
|
| 148 |
+
"generation_config": null,
|
| 149 |
+
"tuner_backend": "peft",
|
| 150 |
+
"vit_gradient_checkpointing": true,
|
| 151 |
+
"router_aux_loss_coef": 0.0,
|
| 152 |
+
"enable_dft_loss": false,
|
| 153 |
+
"enable_channel_loss": false,
|
| 154 |
+
"safe_serialization": true,
|
| 155 |
+
"max_shard_size": "5GB",
|
| 156 |
+
"check_model": true,
|
| 157 |
+
"acc_strategy": "token",
|
| 158 |
+
"train_dataloader_shuffle": true,
|
| 159 |
+
"group_by_length": false,
|
| 160 |
+
"max_epochs": null,
|
| 161 |
+
"aligner_lr": null,
|
| 162 |
+
"vit_lr": 1e-06,
|
| 163 |
+
"use_logits_to_keep": null,
|
| 164 |
+
"ds3_gather_for_generation": true,
|
| 165 |
+
"resume_only_model": false,
|
| 166 |
+
"optimizer": null,
|
| 167 |
+
"loss_type": null,
|
| 168 |
+
"eval_metric": null,
|
| 169 |
+
"callbacks": [],
|
| 170 |
+
"early_stop_interval": null,
|
| 171 |
+
"eval_use_evalscope": false,
|
| 172 |
+
"eval_dataset": [],
|
| 173 |
+
"eval_dataset_args": null,
|
| 174 |
+
"eval_limit": null,
|
| 175 |
+
"eval_generation_config": null,
|
| 176 |
+
"extra_eval_args": null,
|
| 177 |
+
"tuner_type": "full",
|
| 178 |
+
"use_galore": false,
|
| 179 |
+
"galore_target_modules": null,
|
| 180 |
+
"galore_rank": 128,
|
| 181 |
+
"galore_update_proj_gap": 50,
|
| 182 |
+
"galore_scale": 1.0,
|
| 183 |
+
"galore_proj_type": "std",
|
| 184 |
+
"galore_optim_per_parameter": false,
|
| 185 |
+
"galore_with_embedding": false,
|
| 186 |
+
"galore_quantization": false,
|
| 187 |
+
"galore_proj_quant": false,
|
| 188 |
+
"galore_proj_bits": 4,
|
| 189 |
+
"galore_proj_group_size": 256,
|
| 190 |
+
"galore_cos_threshold": 0.4,
|
| 191 |
+
"galore_gamma_proj": 2,
|
| 192 |
+
"galore_queue_size": 5,
|
| 193 |
+
"lisa_activated_layers": 0,
|
| 194 |
+
"lisa_step_interval": 20,
|
| 195 |
+
"use_flash_ckpt": false,
|
| 196 |
+
"use_ray": false,
|
| 197 |
+
"ray_exp_name": null,
|
| 198 |
+
"device_groups": null,
|
| 199 |
+
"model": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_qwen3_5vit_stock_baseline_a800/v0-20260529-170602/checkpoint-2181",
|
| 200 |
+
"model_type": "llava_qwen3_5vit_qwen3",
|
| 201 |
+
"model_revision": null,
|
| 202 |
+
"task_type": "causal_lm",
|
| 203 |
+
"torch_dtype": "bfloat16",
|
| 204 |
+
"attn_impl": "flash_attn",
|
| 205 |
+
"experts_impl": null,
|
| 206 |
+
"new_special_tokens": [],
|
| 207 |
+
"num_labels": null,
|
| 208 |
+
"problem_type": null,
|
| 209 |
+
"rope_scaling": null,
|
| 210 |
+
"device_map": null,
|
| 211 |
+
"max_memory": {},
|
| 212 |
+
"max_model_len": null,
|
| 213 |
+
"local_repo_path": null,
|
| 214 |
+
"init_strategy": null,
|
| 215 |
+
"template": "llava_qwen3_5vit_qwen3",
|
| 216 |
+
"system": null,
|
| 217 |
+
"max_length": 16384,
|
| 218 |
+
"truncation_strategy": "delete",
|
| 219 |
+
"max_pixels": null,
|
| 220 |
+
"agent_template": null,
|
| 221 |
+
"norm_bbox": null,
|
| 222 |
+
"use_chat_template": true,
|
| 223 |
+
"padding_side": "right",
|
| 224 |
+
"padding_free": true,
|
| 225 |
+
"loss_scale": "default",
|
| 226 |
+
"sequence_parallel_size": 1,
|
| 227 |
+
"template_backend": "swift",
|
| 228 |
+
"response_prefix": null,
|
| 229 |
+
"enable_thinking": null,
|
| 230 |
+
"add_non_thinking_prefix": true,
|
| 231 |
+
"dataset": [],
|
| 232 |
+
"val_dataset": [],
|
| 233 |
+
"cached_dataset": [
|
| 234 |
+
"/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/image_10pct/train",
|
| 235 |
+
"/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/video_10pct/train"
|
| 236 |
+
],
|
| 237 |
+
"cached_val_dataset": [],
|
| 238 |
+
"split_dataset_ratio": 0.0,
|
| 239 |
+
"dataset_num_proc": 16,
|
| 240 |
+
"load_from_cache_file": false,
|
| 241 |
+
"dataset_shuffle": true,
|
| 242 |
+
"val_dataset_shuffle": false,
|
| 243 |
+
"streaming": false,
|
| 244 |
+
"interleave_prob": null,
|
| 245 |
+
"stopping_strategy": "first_exhausted",
|
| 246 |
+
"shuffle_buffer_size": 1000,
|
| 247 |
+
"download_mode": "reuse_dataset_if_exists",
|
| 248 |
+
"columns": {},
|
| 249 |
+
"strict": false,
|
| 250 |
+
"model_name": null,
|
| 251 |
+
"model_author": null,
|
| 252 |
+
"custom_dataset_info": [],
|
| 253 |
+
"quant_method": null,
|
| 254 |
+
"quant_bits": null,
|
| 255 |
+
"hqq_axis": null,
|
| 256 |
+
"bnb_4bit_compute_dtype": "bfloat16",
|
| 257 |
+
"bnb_4bit_quant_type": "nf4",
|
| 258 |
+
"bnb_4bit_use_double_quant": true,
|
| 259 |
+
"bnb_4bit_quant_storage": null,
|
| 260 |
+
"max_new_tokens": 64,
|
| 261 |
+
"temperature": 0.0,
|
| 262 |
+
"top_k": null,
|
| 263 |
+
"top_p": null,
|
| 264 |
+
"repetition_penalty": null,
|
| 265 |
+
"num_beams": 1,
|
| 266 |
+
"stream": false,
|
| 267 |
+
"stop_words": [],
|
| 268 |
+
"logprobs": false,
|
| 269 |
+
"top_logprobs": null,
|
| 270 |
+
"structured_outputs_regex": null,
|
| 271 |
+
"adapters": [],
|
| 272 |
+
"external_plugins": [
|
| 273 |
+
"ms-swift/video_mllm/model_plugin_qwen3_5vit.py",
|
| 274 |
+
"ms-swift/video_mllm/dataset_plugin.py"
|
| 275 |
+
],
|
| 276 |
+
"custom_register_path": [],
|
| 277 |
+
"model_kwargs": {},
|
| 278 |
+
"load_args": false,
|
| 279 |
+
"load_data_args": false,
|
| 280 |
+
"packing": true,
|
| 281 |
+
"packing_length": 16384,
|
| 282 |
+
"packing_num_proc": 1,
|
| 283 |
+
"lazy_tokenize": false,
|
| 284 |
+
"use_hf": true,
|
| 285 |
+
"ignore_args_error": false,
|
| 286 |
+
"use_swift_lora": false,
|
| 287 |
+
"freeze_parameters": [],
|
| 288 |
+
"freeze_parameters_regex": null,
|
| 289 |
+
"freeze_parameters_ratio": 0.0,
|
| 290 |
+
"trainable_parameters": [
|
| 291 |
+
"model.multi_modal_projector"
|
| 292 |
+
],
|
| 293 |
+
"trainable_parameters_regex": null,
|
| 294 |
+
"freeze_llm": false,
|
| 295 |
+
"freeze_vit": false,
|
| 296 |
+
"freeze_aligner": false,
|
| 297 |
+
"target_modules": [
|
| 298 |
+
"all-linear"
|
| 299 |
+
],
|
| 300 |
+
"target_regex": null,
|
| 301 |
+
"target_parameters": null,
|
| 302 |
+
"modules_to_save": [],
|
| 303 |
+
"lora_rank": 8,
|
| 304 |
+
"lora_alpha": 32,
|
| 305 |
+
"lora_dropout": 0.05,
|
| 306 |
+
"lora_bias": "none",
|
| 307 |
+
"lora_dtype": null,
|
| 308 |
+
"lorap_lr_ratio": null,
|
| 309 |
+
"use_rslora": false,
|
| 310 |
+
"use_dora": false,
|
| 311 |
+
"lora_ga_batch_size": 2,
|
| 312 |
+
"lora_ga_iters": 2,
|
| 313 |
+
"lora_ga_max_length": 1024,
|
| 314 |
+
"lora_ga_direction": "ArB2r",
|
| 315 |
+
"lora_ga_scale": "stable",
|
| 316 |
+
"lora_ga_stable_gamma": 16,
|
| 317 |
+
"init_weights": true,
|
| 318 |
+
"fourier_n_frequency": 2000,
|
| 319 |
+
"fourier_scaling": 300.0,
|
| 320 |
+
"boft_block_size": 4,
|
| 321 |
+
"boft_block_num": 0,
|
| 322 |
+
"boft_n_butterfly_factor": 1,
|
| 323 |
+
"boft_dropout": 0.0,
|
| 324 |
+
"vera_rank": 256,
|
| 325 |
+
"vera_projection_prng_key": 0,
|
| 326 |
+
"vera_dropout": 0.0,
|
| 327 |
+
"vera_d_initial": 0.1,
|
| 328 |
+
"adapter_act": "gelu",
|
| 329 |
+
"adapter_length": 128,
|
| 330 |
+
"adalora_target_r": 8,
|
| 331 |
+
"adalora_init_r": 12,
|
| 332 |
+
"adalora_tinit": 0,
|
| 333 |
+
"adalora_tfinal": 0,
|
| 334 |
+
"adalora_deltaT": 1,
|
| 335 |
+
"adalora_beta1": 0.85,
|
| 336 |
+
"adalora_beta2": 0.85,
|
| 337 |
+
"adalora_orth_reg_weight": 0.5,
|
| 338 |
+
"llamapro_num_new_blocks": 4,
|
| 339 |
+
"llamapro_num_groups": null,
|
| 340 |
+
"reft_layer_key": null,
|
| 341 |
+
"reft_layers": null,
|
| 342 |
+
"reft_rank": 4,
|
| 343 |
+
"reft_intervention_type": "LoreftIntervention",
|
| 344 |
+
"reft_args": null,
|
| 345 |
+
"swanlab_token": null,
|
| 346 |
+
"swanlab_project": "ms-swift",
|
| 347 |
+
"swanlab_workspace": null,
|
| 348 |
+
"swanlab_exp_name": null,
|
| 349 |
+
"swanlab_notification_method": null,
|
| 350 |
+
"swanlab_webhook_url": null,
|
| 351 |
+
"swanlab_secret": null,
|
| 352 |
+
"swanlab_sender_email": null,
|
| 353 |
+
"swanlab_receiver_email": null,
|
| 354 |
+
"swanlab_smtp_server": null,
|
| 355 |
+
"swanlab_smtp_port": null,
|
| 356 |
+
"swanlab_email_language": "zh",
|
| 357 |
+
"swanlab_mode": "cloud",
|
| 358 |
+
"add_version": true,
|
| 359 |
+
"create_checkpoint_symlink": false,
|
| 360 |
+
"zero_hpz_partition_size": null,
|
| 361 |
+
"deepspeed_autotp_size": null,
|
| 362 |
+
"swift_version": "4.1.1",
|
| 363 |
+
"ckpt_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_qwen3_5vit_stock_baseline_a800/v0-20260529-170602/checkpoint-2181",
|
| 364 |
+
"rank": 0,
|
| 365 |
+
"global_world_size": 8,
|
| 366 |
+
"local_world_size": 8,
|
| 367 |
+
"model_suffix": "checkpoint-2181",
|
| 368 |
+
"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)",
|
| 369 |
+
"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'])",
|
| 370 |
+
"model_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_qwen3_5vit_stock_baseline_a800/v0-20260529-170602/checkpoint-2181",
|
| 371 |
+
"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='')",
|
| 372 |
+
"_val_dataset_exists": false,
|
| 373 |
+
"hub": "<class 'swift.hub.hub.HFHub'>",
|
| 374 |
+
"evaluation_strategy": "steps",
|
| 375 |
+
"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)"
|
| 376 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/chat_template.jinja
ADDED
|
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{%- if tools %}
|
| 2 |
+
{{- '<|im_start|>system\n' }}
|
| 3 |
+
{%- if messages[0].role == 'system' %}
|
| 4 |
+
{{- messages[0].content + '\n\n' }}
|
| 5 |
+
{%- endif %}
|
| 6 |
+
{{- "# 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>" }}
|
| 7 |
+
{%- for tool in tools %}
|
| 8 |
+
{{- "\n" }}
|
| 9 |
+
{{- tool | tojson }}
|
| 10 |
+
{%- endfor %}
|
| 11 |
+
{{- "\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" }}
|
| 12 |
+
{%- else %}
|
| 13 |
+
{%- if messages[0].role == 'system' %}
|
| 14 |
+
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
| 15 |
+
{%- endif %}
|
| 16 |
+
{%- endif %}
|
| 17 |
+
{%- for message in messages %}
|
| 18 |
+
{%- if message.content is string %}
|
| 19 |
+
{%- set content = message.content %}
|
| 20 |
+
{%- else %}
|
| 21 |
+
{%- set content = '' %}
|
| 22 |
+
{%- endif %}
|
| 23 |
+
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
| 24 |
+
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
| 25 |
+
{%- elif message.role == "assistant" %}
|
| 26 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 27 |
+
{%- if message.tool_calls %}
|
| 28 |
+
{%- for tool_call in message.tool_calls %}
|
| 29 |
+
{%- if (loop.first and content) or (not loop.first) %}
|
| 30 |
+
{{- '\n' }}
|
| 31 |
+
{%- endif %}
|
| 32 |
+
{%- if tool_call.function %}
|
| 33 |
+
{%- set tool_call = tool_call.function %}
|
| 34 |
+
{%- endif %}
|
| 35 |
+
{{- '<tool_call>\n{"name": "' }}
|
| 36 |
+
{{- tool_call.name }}
|
| 37 |
+
{{- '", "arguments": ' }}
|
| 38 |
+
{%- if tool_call.arguments is string %}
|
| 39 |
+
{{- tool_call.arguments }}
|
| 40 |
+
{%- else %}
|
| 41 |
+
{{- tool_call.arguments | tojson }}
|
| 42 |
+
{%- endif %}
|
| 43 |
+
{{- '}\n</tool_call>' }}
|
| 44 |
+
{%- endfor %}
|
| 45 |
+
{%- endif %}
|
| 46 |
+
{{- '<|im_end|>\n' }}
|
| 47 |
+
{%- elif message.role == "tool" %}
|
| 48 |
+
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
| 49 |
+
{{- '<|im_start|>user' }}
|
| 50 |
+
{%- endif %}
|
| 51 |
+
{{- '\n<tool_response>\n' }}
|
| 52 |
+
{{- content }}
|
| 53 |
+
{{- '\n</tool_response>' }}
|
| 54 |
+
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
| 55 |
+
{{- '<|im_end|>\n' }}
|
| 56 |
+
{%- endif %}
|
| 57 |
+
{%- endif %}
|
| 58 |
+
{%- endfor %}
|
| 59 |
+
{%- if add_generation_prompt %}
|
| 60 |
+
{{- '<|im_start|>assistant\n' }}
|
| 61 |
+
{%- endif %}
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/chat_template.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"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 %}"
|
| 3 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/config.json
ADDED
|
@@ -0,0 +1,262 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"LlavaQwen3_5ViTForConditionalGeneration"
|
| 4 |
+
],
|
| 5 |
+
"bos_token_id": null,
|
| 6 |
+
"dtype": "bfloat16",
|
| 7 |
+
"eos_token_id": 151645,
|
| 8 |
+
"hidden_size": 2560,
|
| 9 |
+
"image_grid_pinpoints": [
|
| 10 |
+
[
|
| 11 |
+
384,
|
| 12 |
+
384
|
| 13 |
+
],
|
| 14 |
+
[
|
| 15 |
+
384,
|
| 16 |
+
768
|
| 17 |
+
],
|
| 18 |
+
[
|
| 19 |
+
384,
|
| 20 |
+
1152
|
| 21 |
+
],
|
| 22 |
+
[
|
| 23 |
+
384,
|
| 24 |
+
1536
|
| 25 |
+
],
|
| 26 |
+
[
|
| 27 |
+
384,
|
| 28 |
+
1920
|
| 29 |
+
],
|
| 30 |
+
[
|
| 31 |
+
384,
|
| 32 |
+
2304
|
| 33 |
+
],
|
| 34 |
+
[
|
| 35 |
+
768,
|
| 36 |
+
384
|
| 37 |
+
],
|
| 38 |
+
[
|
| 39 |
+
768,
|
| 40 |
+
768
|
| 41 |
+
],
|
| 42 |
+
[
|
| 43 |
+
768,
|
| 44 |
+
1152
|
| 45 |
+
],
|
| 46 |
+
[
|
| 47 |
+
768,
|
| 48 |
+
1536
|
| 49 |
+
],
|
| 50 |
+
[
|
| 51 |
+
768,
|
| 52 |
+
1920
|
| 53 |
+
],
|
| 54 |
+
[
|
| 55 |
+
768,
|
| 56 |
+
2304
|
| 57 |
+
],
|
| 58 |
+
[
|
| 59 |
+
1152,
|
| 60 |
+
384
|
| 61 |
+
],
|
| 62 |
+
[
|
| 63 |
+
1152,
|
| 64 |
+
768
|
| 65 |
+
],
|
| 66 |
+
[
|
| 67 |
+
1152,
|
| 68 |
+
1152
|
| 69 |
+
],
|
| 70 |
+
[
|
| 71 |
+
1152,
|
| 72 |
+
1536
|
| 73 |
+
],
|
| 74 |
+
[
|
| 75 |
+
1152,
|
| 76 |
+
1920
|
| 77 |
+
],
|
| 78 |
+
[
|
| 79 |
+
1152,
|
| 80 |
+
2304
|
| 81 |
+
],
|
| 82 |
+
[
|
| 83 |
+
1536,
|
| 84 |
+
384
|
| 85 |
+
],
|
| 86 |
+
[
|
| 87 |
+
1536,
|
| 88 |
+
768
|
| 89 |
+
],
|
| 90 |
+
[
|
| 91 |
+
1536,
|
| 92 |
+
1152
|
| 93 |
+
],
|
| 94 |
+
[
|
| 95 |
+
1536,
|
| 96 |
+
1536
|
| 97 |
+
],
|
| 98 |
+
[
|
| 99 |
+
1536,
|
| 100 |
+
1920
|
| 101 |
+
],
|
| 102 |
+
[
|
| 103 |
+
1536,
|
| 104 |
+
2304
|
| 105 |
+
],
|
| 106 |
+
[
|
| 107 |
+
1920,
|
| 108 |
+
384
|
| 109 |
+
],
|
| 110 |
+
[
|
| 111 |
+
1920,
|
| 112 |
+
768
|
| 113 |
+
],
|
| 114 |
+
[
|
| 115 |
+
1920,
|
| 116 |
+
1152
|
| 117 |
+
],
|
| 118 |
+
[
|
| 119 |
+
1920,
|
| 120 |
+
1536
|
| 121 |
+
],
|
| 122 |
+
[
|
| 123 |
+
1920,
|
| 124 |
+
1920
|
| 125 |
+
],
|
| 126 |
+
[
|
| 127 |
+
1920,
|
| 128 |
+
2304
|
| 129 |
+
],
|
| 130 |
+
[
|
| 131 |
+
2304,
|
| 132 |
+
384
|
| 133 |
+
],
|
| 134 |
+
[
|
| 135 |
+
2304,
|
| 136 |
+
768
|
| 137 |
+
],
|
| 138 |
+
[
|
| 139 |
+
2304,
|
| 140 |
+
1152
|
| 141 |
+
],
|
| 142 |
+
[
|
| 143 |
+
2304,
|
| 144 |
+
1536
|
| 145 |
+
],
|
| 146 |
+
[
|
| 147 |
+
2304,
|
| 148 |
+
1920
|
| 149 |
+
],
|
| 150 |
+
[
|
| 151 |
+
2304,
|
| 152 |
+
2304
|
| 153 |
+
]
|
| 154 |
+
],
|
| 155 |
+
"image_token_index": 151669,
|
| 156 |
+
"model_type": "llava_qwen3_5vit_qwen3",
|
| 157 |
+
"multimodal_projector_bias": true,
|
| 158 |
+
"pad_token_id": 151643,
|
| 159 |
+
"projector_hidden_act": "gelu",
|
| 160 |
+
"text_config": {
|
| 161 |
+
"_name_or_path": "/share/m2v_intern_v3/wangjunjie09/model_cache/huggingface/Qwen/Qwen3-4B-Instruct-2507",
|
| 162 |
+
"architectures": [
|
| 163 |
+
"Qwen3ForCausalLM"
|
| 164 |
+
],
|
| 165 |
+
"attention_bias": false,
|
| 166 |
+
"attention_dropout": 0.0,
|
| 167 |
+
"bos_token_id": 151643,
|
| 168 |
+
"dtype": "bfloat16",
|
| 169 |
+
"eos_token_id": 151645,
|
| 170 |
+
"head_dim": 128,
|
| 171 |
+
"hidden_act": "silu",
|
| 172 |
+
"hidden_size": 2560,
|
| 173 |
+
"initializer_range": 0.02,
|
| 174 |
+
"intermediate_size": 9728,
|
| 175 |
+
"layer_types": [
|
| 176 |
+
"full_attention",
|
| 177 |
+
"full_attention",
|
| 178 |
+
"full_attention",
|
| 179 |
+
"full_attention",
|
| 180 |
+
"full_attention",
|
| 181 |
+
"full_attention",
|
| 182 |
+
"full_attention",
|
| 183 |
+
"full_attention",
|
| 184 |
+
"full_attention",
|
| 185 |
+
"full_attention",
|
| 186 |
+
"full_attention",
|
| 187 |
+
"full_attention",
|
| 188 |
+
"full_attention",
|
| 189 |
+
"full_attention",
|
| 190 |
+
"full_attention",
|
| 191 |
+
"full_attention",
|
| 192 |
+
"full_attention",
|
| 193 |
+
"full_attention",
|
| 194 |
+
"full_attention",
|
| 195 |
+
"full_attention",
|
| 196 |
+
"full_attention",
|
| 197 |
+
"full_attention",
|
| 198 |
+
"full_attention",
|
| 199 |
+
"full_attention",
|
| 200 |
+
"full_attention",
|
| 201 |
+
"full_attention",
|
| 202 |
+
"full_attention",
|
| 203 |
+
"full_attention",
|
| 204 |
+
"full_attention",
|
| 205 |
+
"full_attention",
|
| 206 |
+
"full_attention",
|
| 207 |
+
"full_attention",
|
| 208 |
+
"full_attention",
|
| 209 |
+
"full_attention",
|
| 210 |
+
"full_attention",
|
| 211 |
+
"full_attention"
|
| 212 |
+
],
|
| 213 |
+
"max_position_embeddings": 262144,
|
| 214 |
+
"max_window_layers": 36,
|
| 215 |
+
"model_type": "qwen3",
|
| 216 |
+
"num_attention_heads": 32,
|
| 217 |
+
"num_hidden_layers": 36,
|
| 218 |
+
"num_key_value_heads": 8,
|
| 219 |
+
"pad_token_id": 151643,
|
| 220 |
+
"rms_norm_eps": 1e-06,
|
| 221 |
+
"rope_parameters": {
|
| 222 |
+
"rope_theta": 5000000,
|
| 223 |
+
"rope_type": "default"
|
| 224 |
+
},
|
| 225 |
+
"sliding_window": null,
|
| 226 |
+
"tie_word_embeddings": true,
|
| 227 |
+
"use_cache": false,
|
| 228 |
+
"use_sliding_window": false,
|
| 229 |
+
"vocab_size": 151936
|
| 230 |
+
},
|
| 231 |
+
"tie_word_embeddings": true,
|
| 232 |
+
"transformers_version": "5.5.4",
|
| 233 |
+
"use_cache": false,
|
| 234 |
+
"video_token_index": 151670,
|
| 235 |
+
"vision_aspect_ratio": "anyres_max_9",
|
| 236 |
+
"vision_config": {
|
| 237 |
+
"deepstack_visual_indexes": [],
|
| 238 |
+
"depth": 24,
|
| 239 |
+
"dtype": "bfloat16",
|
| 240 |
+
"hidden_act": "gelu_pytorch_tanh",
|
| 241 |
+
"hidden_size": 1024,
|
| 242 |
+
"image_size": 384,
|
| 243 |
+
"in_channels": 3,
|
| 244 |
+
"initializer_range": 0.02,
|
| 245 |
+
"intermediate_size": 4096,
|
| 246 |
+
"model_type": "qwen3_5",
|
| 247 |
+
"num_heads": 16,
|
| 248 |
+
"num_position_embeddings": 2304,
|
| 249 |
+
"out_hidden_size": 2048,
|
| 250 |
+
"patch_size": 16,
|
| 251 |
+
"spatial_merge_size": 2,
|
| 252 |
+
"temporal_patch_size": 2
|
| 253 |
+
},
|
| 254 |
+
"vision_feature_layer": -1,
|
| 255 |
+
"vision_feature_select_strategy": "full",
|
| 256 |
+
"auto_map": {
|
| 257 |
+
"AutoConfig": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTConfig",
|
| 258 |
+
"AutoModel": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
|
| 259 |
+
"AutoModelForCausalLM": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
|
| 260 |
+
"AutoModelForImageTextToText": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration"
|
| 261 |
+
}
|
| 262 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/generation_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"bos_token_id": 151643,
|
| 4 |
+
"eos_token_id": [
|
| 5 |
+
151645,
|
| 6 |
+
151643
|
| 7 |
+
],
|
| 8 |
+
"output_attentions": false,
|
| 9 |
+
"output_hidden_states": false,
|
| 10 |
+
"transformers_version": "5.5.4",
|
| 11 |
+
"use_cache": true
|
| 12 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/latest
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
global_step505
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/model-00001-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:87bf8644b6e1d76637dd2c54e296952c9906224aa34740c30b757af2623e2a3c
|
| 3 |
+
size 4987507408
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/model-00002-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:feb82dce0fbd8cd4e8f27f887a5673c7d519f735e368c5f7fce50663f4317757
|
| 3 |
+
size 4466285248
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/model.safetensors.index.json
ADDED
|
@@ -0,0 +1,705 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metadata": {
|
| 3 |
+
"total_parameters": 4337895424,
|
| 4 |
+
"total_size": 9453703168
|
| 5 |
+
},
|
| 6 |
+
"weight_map": {
|
| 7 |
+
"lm_head.weight": "model-00001-of-00002.safetensors",
|
| 8 |
+
"model.image_newline": "model-00001-of-00002.safetensors",
|
| 9 |
+
"model.language_model.embed_tokens.weight": "model-00001-of-00002.safetensors",
|
| 10 |
+
"model.language_model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 11 |
+
"model.language_model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 12 |
+
"model.language_model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 13 |
+
"model.language_model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 14 |
+
"model.language_model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 15 |
+
"model.language_model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 16 |
+
"model.language_model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 17 |
+
"model.language_model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 18 |
+
"model.language_model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 19 |
+
"model.language_model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 20 |
+
"model.language_model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 21 |
+
"model.language_model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 22 |
+
"model.language_model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 23 |
+
"model.language_model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 24 |
+
"model.language_model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 25 |
+
"model.language_model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 26 |
+
"model.language_model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 27 |
+
"model.language_model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 28 |
+
"model.language_model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 29 |
+
"model.language_model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 30 |
+
"model.language_model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 31 |
+
"model.language_model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 32 |
+
"model.language_model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 33 |
+
"model.language_model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 34 |
+
"model.language_model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 35 |
+
"model.language_model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 36 |
+
"model.language_model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 37 |
+
"model.language_model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 38 |
+
"model.language_model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 39 |
+
"model.language_model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 40 |
+
"model.language_model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 41 |
+
"model.language_model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 42 |
+
"model.language_model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 43 |
+
"model.language_model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 44 |
+
"model.language_model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 45 |
+
"model.language_model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 46 |
+
"model.language_model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 47 |
+
"model.language_model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 48 |
+
"model.language_model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 49 |
+
"model.language_model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 50 |
+
"model.language_model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 51 |
+
"model.language_model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 52 |
+
"model.language_model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 53 |
+
"model.language_model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 54 |
+
"model.language_model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 55 |
+
"model.language_model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 56 |
+
"model.language_model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 57 |
+
"model.language_model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 58 |
+
"model.language_model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 59 |
+
"model.language_model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 60 |
+
"model.language_model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 61 |
+
"model.language_model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 62 |
+
"model.language_model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 63 |
+
"model.language_model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 64 |
+
"model.language_model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 65 |
+
"model.language_model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 66 |
+
"model.language_model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 67 |
+
"model.language_model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 68 |
+
"model.language_model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 69 |
+
"model.language_model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 70 |
+
"model.language_model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 71 |
+
"model.language_model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 72 |
+
"model.language_model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 73 |
+
"model.language_model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 74 |
+
"model.language_model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 75 |
+
"model.language_model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 76 |
+
"model.language_model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 77 |
+
"model.language_model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 78 |
+
"model.language_model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 79 |
+
"model.language_model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 80 |
+
"model.language_model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 81 |
+
"model.language_model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 82 |
+
"model.language_model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 83 |
+
"model.language_model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 84 |
+
"model.language_model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 85 |
+
"model.language_model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 86 |
+
"model.language_model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 87 |
+
"model.language_model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 88 |
+
"model.language_model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 89 |
+
"model.language_model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 90 |
+
"model.language_model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 91 |
+
"model.language_model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 92 |
+
"model.language_model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 93 |
+
"model.language_model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 94 |
+
"model.language_model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 95 |
+
"model.language_model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 96 |
+
"model.language_model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 97 |
+
"model.language_model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 98 |
+
"model.language_model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 99 |
+
"model.language_model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 100 |
+
"model.language_model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 101 |
+
"model.language_model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 102 |
+
"model.language_model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 103 |
+
"model.language_model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 104 |
+
"model.language_model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 105 |
+
"model.language_model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 106 |
+
"model.language_model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 107 |
+
"model.language_model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 108 |
+
"model.language_model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 109 |
+
"model.language_model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 110 |
+
"model.language_model.layers.17.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 111 |
+
"model.language_model.layers.17.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 112 |
+
"model.language_model.layers.17.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 113 |
+
"model.language_model.layers.17.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 114 |
+
"model.language_model.layers.17.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 115 |
+
"model.language_model.layers.17.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 116 |
+
"model.language_model.layers.17.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 117 |
+
"model.language_model.layers.17.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 118 |
+
"model.language_model.layers.17.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 119 |
+
"model.language_model.layers.17.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 120 |
+
"model.language_model.layers.18.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 121 |
+
"model.language_model.layers.18.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 122 |
+
"model.language_model.layers.18.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 123 |
+
"model.language_model.layers.18.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 124 |
+
"model.language_model.layers.18.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 125 |
+
"model.language_model.layers.18.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 126 |
+
"model.language_model.layers.18.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 127 |
+
"model.language_model.layers.18.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 128 |
+
"model.language_model.layers.18.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 129 |
+
"model.language_model.layers.18.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 130 |
+
"model.language_model.layers.18.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 131 |
+
"model.language_model.layers.19.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 132 |
+
"model.language_model.layers.19.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 133 |
+
"model.language_model.layers.19.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 134 |
+
"model.language_model.layers.19.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 135 |
+
"model.language_model.layers.19.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 136 |
+
"model.language_model.layers.19.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 137 |
+
"model.language_model.layers.19.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 138 |
+
"model.language_model.layers.19.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 139 |
+
"model.language_model.layers.19.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 140 |
+
"model.language_model.layers.19.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 141 |
+
"model.language_model.layers.19.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 142 |
+
"model.language_model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 143 |
+
"model.language_model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 144 |
+
"model.language_model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 145 |
+
"model.language_model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 146 |
+
"model.language_model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 147 |
+
"model.language_model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 148 |
+
"model.language_model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 149 |
+
"model.language_model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 150 |
+
"model.language_model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 151 |
+
"model.language_model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 152 |
+
"model.language_model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 153 |
+
"model.language_model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 154 |
+
"model.language_model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 155 |
+
"model.language_model.layers.20.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 156 |
+
"model.language_model.layers.20.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 157 |
+
"model.language_model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 158 |
+
"model.language_model.layers.20.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 159 |
+
"model.language_model.layers.20.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 160 |
+
"model.language_model.layers.20.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 161 |
+
"model.language_model.layers.20.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 162 |
+
"model.language_model.layers.20.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 163 |
+
"model.language_model.layers.20.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 164 |
+
"model.language_model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 165 |
+
"model.language_model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 166 |
+
"model.language_model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 167 |
+
"model.language_model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 168 |
+
"model.language_model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 169 |
+
"model.language_model.layers.21.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 170 |
+
"model.language_model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 171 |
+
"model.language_model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 172 |
+
"model.language_model.layers.21.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 173 |
+
"model.language_model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 174 |
+
"model.language_model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 175 |
+
"model.language_model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 176 |
+
"model.language_model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 177 |
+
"model.language_model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 178 |
+
"model.language_model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 179 |
+
"model.language_model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 180 |
+
"model.language_model.layers.22.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 181 |
+
"model.language_model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 182 |
+
"model.language_model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 183 |
+
"model.language_model.layers.22.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 184 |
+
"model.language_model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 185 |
+
"model.language_model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 186 |
+
"model.language_model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 187 |
+
"model.language_model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 188 |
+
"model.language_model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 189 |
+
"model.language_model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 190 |
+
"model.language_model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 191 |
+
"model.language_model.layers.23.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 192 |
+
"model.language_model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 193 |
+
"model.language_model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 194 |
+
"model.language_model.layers.23.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 195 |
+
"model.language_model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 196 |
+
"model.language_model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 197 |
+
"model.language_model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 198 |
+
"model.language_model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 199 |
+
"model.language_model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 200 |
+
"model.language_model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 201 |
+
"model.language_model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 202 |
+
"model.language_model.layers.24.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 203 |
+
"model.language_model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 204 |
+
"model.language_model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 205 |
+
"model.language_model.layers.24.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 206 |
+
"model.language_model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 207 |
+
"model.language_model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 208 |
+
"model.language_model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 209 |
+
"model.language_model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 210 |
+
"model.language_model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 211 |
+
"model.language_model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 212 |
+
"model.language_model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 213 |
+
"model.language_model.layers.25.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 214 |
+
"model.language_model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 215 |
+
"model.language_model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 216 |
+
"model.language_model.layers.25.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 217 |
+
"model.language_model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 218 |
+
"model.language_model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 219 |
+
"model.language_model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 220 |
+
"model.language_model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 221 |
+
"model.language_model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 222 |
+
"model.language_model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 223 |
+
"model.language_model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 224 |
+
"model.language_model.layers.26.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 225 |
+
"model.language_model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 226 |
+
"model.language_model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 227 |
+
"model.language_model.layers.26.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 228 |
+
"model.language_model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 229 |
+
"model.language_model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 230 |
+
"model.language_model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 231 |
+
"model.language_model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 232 |
+
"model.language_model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 233 |
+
"model.language_model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 234 |
+
"model.language_model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 235 |
+
"model.language_model.layers.27.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 236 |
+
"model.language_model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 237 |
+
"model.language_model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 238 |
+
"model.language_model.layers.27.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 239 |
+
"model.language_model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 240 |
+
"model.language_model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 241 |
+
"model.language_model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 242 |
+
"model.language_model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 243 |
+
"model.language_model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 244 |
+
"model.language_model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 245 |
+
"model.language_model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 246 |
+
"model.language_model.layers.28.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 247 |
+
"model.language_model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 248 |
+
"model.language_model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 249 |
+
"model.language_model.layers.28.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 250 |
+
"model.language_model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 251 |
+
"model.language_model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 252 |
+
"model.language_model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 253 |
+
"model.language_model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 254 |
+
"model.language_model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 255 |
+
"model.language_model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 256 |
+
"model.language_model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 257 |
+
"model.language_model.layers.29.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 258 |
+
"model.language_model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 259 |
+
"model.language_model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 260 |
+
"model.language_model.layers.29.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 261 |
+
"model.language_model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 262 |
+
"model.language_model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 263 |
+
"model.language_model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 264 |
+
"model.language_model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 265 |
+
"model.language_model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 266 |
+
"model.language_model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 267 |
+
"model.language_model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 268 |
+
"model.language_model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 269 |
+
"model.language_model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 270 |
+
"model.language_model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 271 |
+
"model.language_model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 272 |
+
"model.language_model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 273 |
+
"model.language_model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 274 |
+
"model.language_model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 275 |
+
"model.language_model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 276 |
+
"model.language_model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 277 |
+
"model.language_model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 278 |
+
"model.language_model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 279 |
+
"model.language_model.layers.30.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 280 |
+
"model.language_model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 281 |
+
"model.language_model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 282 |
+
"model.language_model.layers.30.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 283 |
+
"model.language_model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 284 |
+
"model.language_model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 285 |
+
"model.language_model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 286 |
+
"model.language_model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 287 |
+
"model.language_model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 288 |
+
"model.language_model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 289 |
+
"model.language_model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 290 |
+
"model.language_model.layers.31.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 291 |
+
"model.language_model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 292 |
+
"model.language_model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 293 |
+
"model.language_model.layers.31.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 294 |
+
"model.language_model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 295 |
+
"model.language_model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 296 |
+
"model.language_model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 297 |
+
"model.language_model.layers.32.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 298 |
+
"model.language_model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 299 |
+
"model.language_model.layers.32.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 300 |
+
"model.language_model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 301 |
+
"model.language_model.layers.32.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 302 |
+
"model.language_model.layers.32.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 303 |
+
"model.language_model.layers.32.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 304 |
+
"model.language_model.layers.32.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 305 |
+
"model.language_model.layers.32.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 306 |
+
"model.language_model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 307 |
+
"model.language_model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 308 |
+
"model.language_model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 309 |
+
"model.language_model.layers.33.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 310 |
+
"model.language_model.layers.33.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 311 |
+
"model.language_model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 312 |
+
"model.language_model.layers.33.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 313 |
+
"model.language_model.layers.33.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 314 |
+
"model.language_model.layers.33.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 315 |
+
"model.language_model.layers.33.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 316 |
+
"model.language_model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 317 |
+
"model.language_model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 318 |
+
"model.language_model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 319 |
+
"model.language_model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 320 |
+
"model.language_model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 321 |
+
"model.language_model.layers.34.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 322 |
+
"model.language_model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 323 |
+
"model.language_model.layers.34.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 324 |
+
"model.language_model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 325 |
+
"model.language_model.layers.34.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 326 |
+
"model.language_model.layers.34.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 327 |
+
"model.language_model.layers.34.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 328 |
+
"model.language_model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 329 |
+
"model.language_model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 330 |
+
"model.language_model.layers.35.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 331 |
+
"model.language_model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 332 |
+
"model.language_model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 333 |
+
"model.language_model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 334 |
+
"model.language_model.layers.35.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 335 |
+
"model.language_model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 336 |
+
"model.language_model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 337 |
+
"model.language_model.layers.35.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 338 |
+
"model.language_model.layers.35.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 339 |
+
"model.language_model.layers.35.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 340 |
+
"model.language_model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 341 |
+
"model.language_model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 342 |
+
"model.language_model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 343 |
+
"model.language_model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 344 |
+
"model.language_model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 345 |
+
"model.language_model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 346 |
+
"model.language_model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 347 |
+
"model.language_model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 348 |
+
"model.language_model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 349 |
+
"model.language_model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 350 |
+
"model.language_model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 351 |
+
"model.language_model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 352 |
+
"model.language_model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 353 |
+
"model.language_model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 354 |
+
"model.language_model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 355 |
+
"model.language_model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 356 |
+
"model.language_model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 357 |
+
"model.language_model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 358 |
+
"model.language_model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 359 |
+
"model.language_model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 360 |
+
"model.language_model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 361 |
+
"model.language_model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 362 |
+
"model.language_model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 363 |
+
"model.language_model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 364 |
+
"model.language_model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 365 |
+
"model.language_model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 366 |
+
"model.language_model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 367 |
+
"model.language_model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 368 |
+
"model.language_model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 369 |
+
"model.language_model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 370 |
+
"model.language_model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 371 |
+
"model.language_model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 372 |
+
"model.language_model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 373 |
+
"model.language_model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 374 |
+
"model.language_model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 375 |
+
"model.language_model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 376 |
+
"model.language_model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 377 |
+
"model.language_model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 378 |
+
"model.language_model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 379 |
+
"model.language_model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 380 |
+
"model.language_model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 381 |
+
"model.language_model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 382 |
+
"model.language_model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 383 |
+
"model.language_model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 384 |
+
"model.language_model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 385 |
+
"model.language_model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 386 |
+
"model.language_model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 387 |
+
"model.language_model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 388 |
+
"model.language_model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 389 |
+
"model.language_model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 390 |
+
"model.language_model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 391 |
+
"model.language_model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 392 |
+
"model.language_model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 393 |
+
"model.language_model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 394 |
+
"model.language_model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 395 |
+
"model.language_model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 396 |
+
"model.language_model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 397 |
+
"model.language_model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 398 |
+
"model.language_model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 399 |
+
"model.language_model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 400 |
+
"model.language_model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 401 |
+
"model.language_model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 402 |
+
"model.language_model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 403 |
+
"model.language_model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 404 |
+
"model.language_model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 405 |
+
"model.language_model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 406 |
+
"model.language_model.norm.weight": "model-00002-of-00002.safetensors",
|
| 407 |
+
"model.multi_modal_projector.linear_1.bias": "model-00002-of-00002.safetensors",
|
| 408 |
+
"model.multi_modal_projector.linear_1.weight": "model-00002-of-00002.safetensors",
|
| 409 |
+
"model.multi_modal_projector.linear_2.bias": "model-00002-of-00002.safetensors",
|
| 410 |
+
"model.multi_modal_projector.linear_2.weight": "model-00002-of-00002.safetensors",
|
| 411 |
+
"model.vision_tower.vision.blocks.0.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 412 |
+
"model.vision_tower.vision.blocks.0.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 413 |
+
"model.vision_tower.vision.blocks.0.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 414 |
+
"model.vision_tower.vision.blocks.0.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 415 |
+
"model.vision_tower.vision.blocks.0.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 416 |
+
"model.vision_tower.vision.blocks.0.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 417 |
+
"model.vision_tower.vision.blocks.0.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 418 |
+
"model.vision_tower.vision.blocks.0.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 419 |
+
"model.vision_tower.vision.blocks.0.norm1.bias": "model-00002-of-00002.safetensors",
|
| 420 |
+
"model.vision_tower.vision.blocks.0.norm1.weight": "model-00002-of-00002.safetensors",
|
| 421 |
+
"model.vision_tower.vision.blocks.0.norm2.bias": "model-00002-of-00002.safetensors",
|
| 422 |
+
"model.vision_tower.vision.blocks.0.norm2.weight": "model-00002-of-00002.safetensors",
|
| 423 |
+
"model.vision_tower.vision.blocks.1.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 424 |
+
"model.vision_tower.vision.blocks.1.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 425 |
+
"model.vision_tower.vision.blocks.1.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 426 |
+
"model.vision_tower.vision.blocks.1.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 427 |
+
"model.vision_tower.vision.blocks.1.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 428 |
+
"model.vision_tower.vision.blocks.1.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 429 |
+
"model.vision_tower.vision.blocks.1.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 430 |
+
"model.vision_tower.vision.blocks.1.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 431 |
+
"model.vision_tower.vision.blocks.1.norm1.bias": "model-00002-of-00002.safetensors",
|
| 432 |
+
"model.vision_tower.vision.blocks.1.norm1.weight": "model-00002-of-00002.safetensors",
|
| 433 |
+
"model.vision_tower.vision.blocks.1.norm2.bias": "model-00002-of-00002.safetensors",
|
| 434 |
+
"model.vision_tower.vision.blocks.1.norm2.weight": "model-00002-of-00002.safetensors",
|
| 435 |
+
"model.vision_tower.vision.blocks.10.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 436 |
+
"model.vision_tower.vision.blocks.10.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 437 |
+
"model.vision_tower.vision.blocks.10.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 438 |
+
"model.vision_tower.vision.blocks.10.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 439 |
+
"model.vision_tower.vision.blocks.10.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 440 |
+
"model.vision_tower.vision.blocks.10.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 441 |
+
"model.vision_tower.vision.blocks.10.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 442 |
+
"model.vision_tower.vision.blocks.10.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 443 |
+
"model.vision_tower.vision.blocks.10.norm1.bias": "model-00002-of-00002.safetensors",
|
| 444 |
+
"model.vision_tower.vision.blocks.10.norm1.weight": "model-00002-of-00002.safetensors",
|
| 445 |
+
"model.vision_tower.vision.blocks.10.norm2.bias": "model-00002-of-00002.safetensors",
|
| 446 |
+
"model.vision_tower.vision.blocks.10.norm2.weight": "model-00002-of-00002.safetensors",
|
| 447 |
+
"model.vision_tower.vision.blocks.11.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 448 |
+
"model.vision_tower.vision.blocks.11.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 449 |
+
"model.vision_tower.vision.blocks.11.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 450 |
+
"model.vision_tower.vision.blocks.11.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 451 |
+
"model.vision_tower.vision.blocks.11.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 452 |
+
"model.vision_tower.vision.blocks.11.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 453 |
+
"model.vision_tower.vision.blocks.11.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 454 |
+
"model.vision_tower.vision.blocks.11.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 455 |
+
"model.vision_tower.vision.blocks.11.norm1.bias": "model-00002-of-00002.safetensors",
|
| 456 |
+
"model.vision_tower.vision.blocks.11.norm1.weight": "model-00002-of-00002.safetensors",
|
| 457 |
+
"model.vision_tower.vision.blocks.11.norm2.bias": "model-00002-of-00002.safetensors",
|
| 458 |
+
"model.vision_tower.vision.blocks.11.norm2.weight": "model-00002-of-00002.safetensors",
|
| 459 |
+
"model.vision_tower.vision.blocks.12.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 460 |
+
"model.vision_tower.vision.blocks.12.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 461 |
+
"model.vision_tower.vision.blocks.12.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 462 |
+
"model.vision_tower.vision.blocks.12.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 463 |
+
"model.vision_tower.vision.blocks.12.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 464 |
+
"model.vision_tower.vision.blocks.12.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 465 |
+
"model.vision_tower.vision.blocks.12.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 466 |
+
"model.vision_tower.vision.blocks.12.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 467 |
+
"model.vision_tower.vision.blocks.12.norm1.bias": "model-00002-of-00002.safetensors",
|
| 468 |
+
"model.vision_tower.vision.blocks.12.norm1.weight": "model-00002-of-00002.safetensors",
|
| 469 |
+
"model.vision_tower.vision.blocks.12.norm2.bias": "model-00002-of-00002.safetensors",
|
| 470 |
+
"model.vision_tower.vision.blocks.12.norm2.weight": "model-00002-of-00002.safetensors",
|
| 471 |
+
"model.vision_tower.vision.blocks.13.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 472 |
+
"model.vision_tower.vision.blocks.13.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 473 |
+
"model.vision_tower.vision.blocks.13.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 474 |
+
"model.vision_tower.vision.blocks.13.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 475 |
+
"model.vision_tower.vision.blocks.13.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 476 |
+
"model.vision_tower.vision.blocks.13.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 477 |
+
"model.vision_tower.vision.blocks.13.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 478 |
+
"model.vision_tower.vision.blocks.13.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 479 |
+
"model.vision_tower.vision.blocks.13.norm1.bias": "model-00002-of-00002.safetensors",
|
| 480 |
+
"model.vision_tower.vision.blocks.13.norm1.weight": "model-00002-of-00002.safetensors",
|
| 481 |
+
"model.vision_tower.vision.blocks.13.norm2.bias": "model-00002-of-00002.safetensors",
|
| 482 |
+
"model.vision_tower.vision.blocks.13.norm2.weight": "model-00002-of-00002.safetensors",
|
| 483 |
+
"model.vision_tower.vision.blocks.14.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 484 |
+
"model.vision_tower.vision.blocks.14.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 485 |
+
"model.vision_tower.vision.blocks.14.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 486 |
+
"model.vision_tower.vision.blocks.14.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 487 |
+
"model.vision_tower.vision.blocks.14.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 488 |
+
"model.vision_tower.vision.blocks.14.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 489 |
+
"model.vision_tower.vision.blocks.14.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 490 |
+
"model.vision_tower.vision.blocks.14.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 491 |
+
"model.vision_tower.vision.blocks.14.norm1.bias": "model-00002-of-00002.safetensors",
|
| 492 |
+
"model.vision_tower.vision.blocks.14.norm1.weight": "model-00002-of-00002.safetensors",
|
| 493 |
+
"model.vision_tower.vision.blocks.14.norm2.bias": "model-00002-of-00002.safetensors",
|
| 494 |
+
"model.vision_tower.vision.blocks.14.norm2.weight": "model-00002-of-00002.safetensors",
|
| 495 |
+
"model.vision_tower.vision.blocks.15.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 496 |
+
"model.vision_tower.vision.blocks.15.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 497 |
+
"model.vision_tower.vision.blocks.15.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 498 |
+
"model.vision_tower.vision.blocks.15.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 499 |
+
"model.vision_tower.vision.blocks.15.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 500 |
+
"model.vision_tower.vision.blocks.15.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 501 |
+
"model.vision_tower.vision.blocks.15.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 502 |
+
"model.vision_tower.vision.blocks.15.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 503 |
+
"model.vision_tower.vision.blocks.15.norm1.bias": "model-00002-of-00002.safetensors",
|
| 504 |
+
"model.vision_tower.vision.blocks.15.norm1.weight": "model-00002-of-00002.safetensors",
|
| 505 |
+
"model.vision_tower.vision.blocks.15.norm2.bias": "model-00002-of-00002.safetensors",
|
| 506 |
+
"model.vision_tower.vision.blocks.15.norm2.weight": "model-00002-of-00002.safetensors",
|
| 507 |
+
"model.vision_tower.vision.blocks.16.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 508 |
+
"model.vision_tower.vision.blocks.16.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 509 |
+
"model.vision_tower.vision.blocks.16.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 510 |
+
"model.vision_tower.vision.blocks.16.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 511 |
+
"model.vision_tower.vision.blocks.16.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 512 |
+
"model.vision_tower.vision.blocks.16.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 513 |
+
"model.vision_tower.vision.blocks.16.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 514 |
+
"model.vision_tower.vision.blocks.16.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 515 |
+
"model.vision_tower.vision.blocks.16.norm1.bias": "model-00002-of-00002.safetensors",
|
| 516 |
+
"model.vision_tower.vision.blocks.16.norm1.weight": "model-00002-of-00002.safetensors",
|
| 517 |
+
"model.vision_tower.vision.blocks.16.norm2.bias": "model-00002-of-00002.safetensors",
|
| 518 |
+
"model.vision_tower.vision.blocks.16.norm2.weight": "model-00002-of-00002.safetensors",
|
| 519 |
+
"model.vision_tower.vision.blocks.17.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 520 |
+
"model.vision_tower.vision.blocks.17.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 521 |
+
"model.vision_tower.vision.blocks.17.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 522 |
+
"model.vision_tower.vision.blocks.17.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 523 |
+
"model.vision_tower.vision.blocks.17.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 524 |
+
"model.vision_tower.vision.blocks.17.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 525 |
+
"model.vision_tower.vision.blocks.17.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 526 |
+
"model.vision_tower.vision.blocks.17.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 527 |
+
"model.vision_tower.vision.blocks.17.norm1.bias": "model-00002-of-00002.safetensors",
|
| 528 |
+
"model.vision_tower.vision.blocks.17.norm1.weight": "model-00002-of-00002.safetensors",
|
| 529 |
+
"model.vision_tower.vision.blocks.17.norm2.bias": "model-00002-of-00002.safetensors",
|
| 530 |
+
"model.vision_tower.vision.blocks.17.norm2.weight": "model-00002-of-00002.safetensors",
|
| 531 |
+
"model.vision_tower.vision.blocks.18.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 532 |
+
"model.vision_tower.vision.blocks.18.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 533 |
+
"model.vision_tower.vision.blocks.18.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 534 |
+
"model.vision_tower.vision.blocks.18.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 535 |
+
"model.vision_tower.vision.blocks.18.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 536 |
+
"model.vision_tower.vision.blocks.18.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 537 |
+
"model.vision_tower.vision.blocks.18.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 538 |
+
"model.vision_tower.vision.blocks.18.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 539 |
+
"model.vision_tower.vision.blocks.18.norm1.bias": "model-00002-of-00002.safetensors",
|
| 540 |
+
"model.vision_tower.vision.blocks.18.norm1.weight": "model-00002-of-00002.safetensors",
|
| 541 |
+
"model.vision_tower.vision.blocks.18.norm2.bias": "model-00002-of-00002.safetensors",
|
| 542 |
+
"model.vision_tower.vision.blocks.18.norm2.weight": "model-00002-of-00002.safetensors",
|
| 543 |
+
"model.vision_tower.vision.blocks.19.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 544 |
+
"model.vision_tower.vision.blocks.19.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 545 |
+
"model.vision_tower.vision.blocks.19.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 546 |
+
"model.vision_tower.vision.blocks.19.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 547 |
+
"model.vision_tower.vision.blocks.19.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 548 |
+
"model.vision_tower.vision.blocks.19.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 549 |
+
"model.vision_tower.vision.blocks.19.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 550 |
+
"model.vision_tower.vision.blocks.19.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 551 |
+
"model.vision_tower.vision.blocks.19.norm1.bias": "model-00002-of-00002.safetensors",
|
| 552 |
+
"model.vision_tower.vision.blocks.19.norm1.weight": "model-00002-of-00002.safetensors",
|
| 553 |
+
"model.vision_tower.vision.blocks.19.norm2.bias": "model-00002-of-00002.safetensors",
|
| 554 |
+
"model.vision_tower.vision.blocks.19.norm2.weight": "model-00002-of-00002.safetensors",
|
| 555 |
+
"model.vision_tower.vision.blocks.2.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 556 |
+
"model.vision_tower.vision.blocks.2.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 557 |
+
"model.vision_tower.vision.blocks.2.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 558 |
+
"model.vision_tower.vision.blocks.2.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 559 |
+
"model.vision_tower.vision.blocks.2.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 560 |
+
"model.vision_tower.vision.blocks.2.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 561 |
+
"model.vision_tower.vision.blocks.2.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 562 |
+
"model.vision_tower.vision.blocks.2.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 563 |
+
"model.vision_tower.vision.blocks.2.norm1.bias": "model-00002-of-00002.safetensors",
|
| 564 |
+
"model.vision_tower.vision.blocks.2.norm1.weight": "model-00002-of-00002.safetensors",
|
| 565 |
+
"model.vision_tower.vision.blocks.2.norm2.bias": "model-00002-of-00002.safetensors",
|
| 566 |
+
"model.vision_tower.vision.blocks.2.norm2.weight": "model-00002-of-00002.safetensors",
|
| 567 |
+
"model.vision_tower.vision.blocks.20.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 568 |
+
"model.vision_tower.vision.blocks.20.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 569 |
+
"model.vision_tower.vision.blocks.20.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 570 |
+
"model.vision_tower.vision.blocks.20.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 571 |
+
"model.vision_tower.vision.blocks.20.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 572 |
+
"model.vision_tower.vision.blocks.20.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 573 |
+
"model.vision_tower.vision.blocks.20.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 574 |
+
"model.vision_tower.vision.blocks.20.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 575 |
+
"model.vision_tower.vision.blocks.20.norm1.bias": "model-00002-of-00002.safetensors",
|
| 576 |
+
"model.vision_tower.vision.blocks.20.norm1.weight": "model-00002-of-00002.safetensors",
|
| 577 |
+
"model.vision_tower.vision.blocks.20.norm2.bias": "model-00002-of-00002.safetensors",
|
| 578 |
+
"model.vision_tower.vision.blocks.20.norm2.weight": "model-00002-of-00002.safetensors",
|
| 579 |
+
"model.vision_tower.vision.blocks.21.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 580 |
+
"model.vision_tower.vision.blocks.21.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 581 |
+
"model.vision_tower.vision.blocks.21.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 582 |
+
"model.vision_tower.vision.blocks.21.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 583 |
+
"model.vision_tower.vision.blocks.21.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 584 |
+
"model.vision_tower.vision.blocks.21.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 585 |
+
"model.vision_tower.vision.blocks.21.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 586 |
+
"model.vision_tower.vision.blocks.21.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 587 |
+
"model.vision_tower.vision.blocks.21.norm1.bias": "model-00002-of-00002.safetensors",
|
| 588 |
+
"model.vision_tower.vision.blocks.21.norm1.weight": "model-00002-of-00002.safetensors",
|
| 589 |
+
"model.vision_tower.vision.blocks.21.norm2.bias": "model-00002-of-00002.safetensors",
|
| 590 |
+
"model.vision_tower.vision.blocks.21.norm2.weight": "model-00002-of-00002.safetensors",
|
| 591 |
+
"model.vision_tower.vision.blocks.22.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 592 |
+
"model.vision_tower.vision.blocks.22.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 593 |
+
"model.vision_tower.vision.blocks.22.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 594 |
+
"model.vision_tower.vision.blocks.22.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 595 |
+
"model.vision_tower.vision.blocks.22.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 596 |
+
"model.vision_tower.vision.blocks.22.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 597 |
+
"model.vision_tower.vision.blocks.22.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 598 |
+
"model.vision_tower.vision.blocks.22.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 599 |
+
"model.vision_tower.vision.blocks.22.norm1.bias": "model-00002-of-00002.safetensors",
|
| 600 |
+
"model.vision_tower.vision.blocks.22.norm1.weight": "model-00002-of-00002.safetensors",
|
| 601 |
+
"model.vision_tower.vision.blocks.22.norm2.bias": "model-00002-of-00002.safetensors",
|
| 602 |
+
"model.vision_tower.vision.blocks.22.norm2.weight": "model-00002-of-00002.safetensors",
|
| 603 |
+
"model.vision_tower.vision.blocks.23.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 604 |
+
"model.vision_tower.vision.blocks.23.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 605 |
+
"model.vision_tower.vision.blocks.23.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 606 |
+
"model.vision_tower.vision.blocks.23.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 607 |
+
"model.vision_tower.vision.blocks.23.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 608 |
+
"model.vision_tower.vision.blocks.23.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 609 |
+
"model.vision_tower.vision.blocks.23.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 610 |
+
"model.vision_tower.vision.blocks.23.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 611 |
+
"model.vision_tower.vision.blocks.23.norm1.bias": "model-00002-of-00002.safetensors",
|
| 612 |
+
"model.vision_tower.vision.blocks.23.norm1.weight": "model-00002-of-00002.safetensors",
|
| 613 |
+
"model.vision_tower.vision.blocks.23.norm2.bias": "model-00002-of-00002.safetensors",
|
| 614 |
+
"model.vision_tower.vision.blocks.23.norm2.weight": "model-00002-of-00002.safetensors",
|
| 615 |
+
"model.vision_tower.vision.blocks.3.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 616 |
+
"model.vision_tower.vision.blocks.3.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 617 |
+
"model.vision_tower.vision.blocks.3.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 618 |
+
"model.vision_tower.vision.blocks.3.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 619 |
+
"model.vision_tower.vision.blocks.3.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 620 |
+
"model.vision_tower.vision.blocks.3.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 621 |
+
"model.vision_tower.vision.blocks.3.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 622 |
+
"model.vision_tower.vision.blocks.3.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 623 |
+
"model.vision_tower.vision.blocks.3.norm1.bias": "model-00002-of-00002.safetensors",
|
| 624 |
+
"model.vision_tower.vision.blocks.3.norm1.weight": "model-00002-of-00002.safetensors",
|
| 625 |
+
"model.vision_tower.vision.blocks.3.norm2.bias": "model-00002-of-00002.safetensors",
|
| 626 |
+
"model.vision_tower.vision.blocks.3.norm2.weight": "model-00002-of-00002.safetensors",
|
| 627 |
+
"model.vision_tower.vision.blocks.4.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 628 |
+
"model.vision_tower.vision.blocks.4.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 629 |
+
"model.vision_tower.vision.blocks.4.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 630 |
+
"model.vision_tower.vision.blocks.4.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 631 |
+
"model.vision_tower.vision.blocks.4.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 632 |
+
"model.vision_tower.vision.blocks.4.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 633 |
+
"model.vision_tower.vision.blocks.4.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 634 |
+
"model.vision_tower.vision.blocks.4.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 635 |
+
"model.vision_tower.vision.blocks.4.norm1.bias": "model-00002-of-00002.safetensors",
|
| 636 |
+
"model.vision_tower.vision.blocks.4.norm1.weight": "model-00002-of-00002.safetensors",
|
| 637 |
+
"model.vision_tower.vision.blocks.4.norm2.bias": "model-00002-of-00002.safetensors",
|
| 638 |
+
"model.vision_tower.vision.blocks.4.norm2.weight": "model-00002-of-00002.safetensors",
|
| 639 |
+
"model.vision_tower.vision.blocks.5.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 640 |
+
"model.vision_tower.vision.blocks.5.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 641 |
+
"model.vision_tower.vision.blocks.5.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 642 |
+
"model.vision_tower.vision.blocks.5.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 643 |
+
"model.vision_tower.vision.blocks.5.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 644 |
+
"model.vision_tower.vision.blocks.5.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 645 |
+
"model.vision_tower.vision.blocks.5.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 646 |
+
"model.vision_tower.vision.blocks.5.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 647 |
+
"model.vision_tower.vision.blocks.5.norm1.bias": "model-00002-of-00002.safetensors",
|
| 648 |
+
"model.vision_tower.vision.blocks.5.norm1.weight": "model-00002-of-00002.safetensors",
|
| 649 |
+
"model.vision_tower.vision.blocks.5.norm2.bias": "model-00002-of-00002.safetensors",
|
| 650 |
+
"model.vision_tower.vision.blocks.5.norm2.weight": "model-00002-of-00002.safetensors",
|
| 651 |
+
"model.vision_tower.vision.blocks.6.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 652 |
+
"model.vision_tower.vision.blocks.6.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 653 |
+
"model.vision_tower.vision.blocks.6.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 654 |
+
"model.vision_tower.vision.blocks.6.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 655 |
+
"model.vision_tower.vision.blocks.6.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 656 |
+
"model.vision_tower.vision.blocks.6.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 657 |
+
"model.vision_tower.vision.blocks.6.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 658 |
+
"model.vision_tower.vision.blocks.6.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 659 |
+
"model.vision_tower.vision.blocks.6.norm1.bias": "model-00002-of-00002.safetensors",
|
| 660 |
+
"model.vision_tower.vision.blocks.6.norm1.weight": "model-00002-of-00002.safetensors",
|
| 661 |
+
"model.vision_tower.vision.blocks.6.norm2.bias": "model-00002-of-00002.safetensors",
|
| 662 |
+
"model.vision_tower.vision.blocks.6.norm2.weight": "model-00002-of-00002.safetensors",
|
| 663 |
+
"model.vision_tower.vision.blocks.7.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 664 |
+
"model.vision_tower.vision.blocks.7.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 665 |
+
"model.vision_tower.vision.blocks.7.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 666 |
+
"model.vision_tower.vision.blocks.7.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 667 |
+
"model.vision_tower.vision.blocks.7.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 668 |
+
"model.vision_tower.vision.blocks.7.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 669 |
+
"model.vision_tower.vision.blocks.7.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 670 |
+
"model.vision_tower.vision.blocks.7.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 671 |
+
"model.vision_tower.vision.blocks.7.norm1.bias": "model-00002-of-00002.safetensors",
|
| 672 |
+
"model.vision_tower.vision.blocks.7.norm1.weight": "model-00002-of-00002.safetensors",
|
| 673 |
+
"model.vision_tower.vision.blocks.7.norm2.bias": "model-00002-of-00002.safetensors",
|
| 674 |
+
"model.vision_tower.vision.blocks.7.norm2.weight": "model-00002-of-00002.safetensors",
|
| 675 |
+
"model.vision_tower.vision.blocks.8.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 676 |
+
"model.vision_tower.vision.blocks.8.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 677 |
+
"model.vision_tower.vision.blocks.8.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 678 |
+
"model.vision_tower.vision.blocks.8.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 679 |
+
"model.vision_tower.vision.blocks.8.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 680 |
+
"model.vision_tower.vision.blocks.8.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 681 |
+
"model.vision_tower.vision.blocks.8.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 682 |
+
"model.vision_tower.vision.blocks.8.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 683 |
+
"model.vision_tower.vision.blocks.8.norm1.bias": "model-00002-of-00002.safetensors",
|
| 684 |
+
"model.vision_tower.vision.blocks.8.norm1.weight": "model-00002-of-00002.safetensors",
|
| 685 |
+
"model.vision_tower.vision.blocks.8.norm2.bias": "model-00002-of-00002.safetensors",
|
| 686 |
+
"model.vision_tower.vision.blocks.8.norm2.weight": "model-00002-of-00002.safetensors",
|
| 687 |
+
"model.vision_tower.vision.blocks.9.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 688 |
+
"model.vision_tower.vision.blocks.9.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 689 |
+
"model.vision_tower.vision.blocks.9.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 690 |
+
"model.vision_tower.vision.blocks.9.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 691 |
+
"model.vision_tower.vision.blocks.9.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 692 |
+
"model.vision_tower.vision.blocks.9.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 693 |
+
"model.vision_tower.vision.blocks.9.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 694 |
+
"model.vision_tower.vision.blocks.9.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 695 |
+
"model.vision_tower.vision.blocks.9.norm1.bias": "model-00002-of-00002.safetensors",
|
| 696 |
+
"model.vision_tower.vision.blocks.9.norm1.weight": "model-00002-of-00002.safetensors",
|
| 697 |
+
"model.vision_tower.vision.blocks.9.norm2.bias": "model-00002-of-00002.safetensors",
|
| 698 |
+
"model.vision_tower.vision.blocks.9.norm2.weight": "model-00002-of-00002.safetensors",
|
| 699 |
+
"model.vision_tower.vision.final_layernorm.bias": "model-00002-of-00002.safetensors",
|
| 700 |
+
"model.vision_tower.vision.final_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 701 |
+
"model.vision_tower.vision.patch_embed.proj.bias": "model-00002-of-00002.safetensors",
|
| 702 |
+
"model.vision_tower.vision.patch_embed.proj.weight": "model-00002-of-00002.safetensors",
|
| 703 |
+
"model.vision_tower.vision.pos_embed.weight": "model-00002-of-00002.safetensors"
|
| 704 |
+
}
|
| 705 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/modeling_qwen3_5vit_qwen3.py
ADDED
|
@@ -0,0 +1,351 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
modeling_qwen3_5vit_qwen3.py — Qwen3.5 Vision as SigLIP-compat vision_tower in LlavaOnevision.
|
| 3 |
+
|
| 4 |
+
设计：与 `modeling_qwen3vlvit_qwen3.py` 严格同构，仅换 vision backbone 源：
|
| 5 |
+
- Qwen3VLVisionModel → Qwen3_5VisionModel（继承关系：Qwen3_5VisionModel(Qwen3VLVisionModel) 去 DeepStack）
|
| 6 |
+
- Qwen3VLVisionConfig → Qwen3_5VisionConfig（父类用 AttributeError 哨兵屏蔽 deepstack_visual_indexes）
|
| 7 |
+
|
| 8 |
+
其余（Adapter 契约翻译、MLP projector + pre_norm、LlavaOnevision 继承 wire class）与
|
| 9 |
+
Qwen3-VL ViT pipeline 完全一致。两条 pipeline 并存意义：DeepStack ablation 天然实验组。
|
| 10 |
+
|
| 11 |
+
类层级：
|
| 12 |
+
Qwen3_5ViTBackbone(Qwen3_5VisionModel) — 去 merger，保持 NaViT 契约
|
| 13 |
+
Qwen3_5ViTAsSiglipAdapter(nn.Module) — 持有 Backbone，做 SigLIP ↔ NaViT 契约翻译
|
| 14 |
+
|
| 15 |
+
三方对比公平性：定 384×384 AnyRes tile + 同款 projector 骨架 + 同款 Qwen3-1.7B LLM。
|
| 16 |
+
"""
|
| 17 |
+
|
| 18 |
+
import math
|
| 19 |
+
import os
|
| 20 |
+
import sys
|
| 21 |
+
from typing import Optional
|
| 22 |
+
|
| 23 |
+
import torch
|
| 24 |
+
import torch.nn as nn
|
| 25 |
+
import torch.nn.functional as F
|
| 26 |
+
from transformers import (
|
| 27 |
+
AutoConfig,
|
| 28 |
+
AutoModel,
|
| 29 |
+
AutoModelForCausalLM,
|
| 30 |
+
LlavaOnevisionConfig,
|
| 31 |
+
LlavaOnevisionForConditionalGeneration,
|
| 32 |
+
LlavaOnevisionModel,
|
| 33 |
+
LlavaOnevisionPreTrainedModel,
|
| 34 |
+
Qwen3Config,
|
| 35 |
+
)
|
| 36 |
+
from transformers.activations import ACT2FN
|
| 37 |
+
from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
|
| 38 |
+
from transformers.models.qwen3_5.configuration_qwen3_5 import Qwen3_5VisionConfig
|
| 39 |
+
from transformers.models.qwen3_5.modeling_qwen3_5 import Qwen3_5VisionModel
|
| 40 |
+
|
| 41 |
+
# Shared layout-permutation utility lives in declip_qwenvit (single source of
|
| 42 |
+
# truth — same code path runs in declip-training-side qk_cosine reorder).
|
| 43 |
+
# Add VisionEncoder repo root to sys.path so this modeling file is importable
|
| 44 |
+
# even when the package isn't pip-installed (ms-swift integration loads it
|
| 45 |
+
# via dynamic plugin path).
|
| 46 |
+
_REPO_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
|
| 47 |
+
if _REPO_ROOT not in sys.path:
|
| 48 |
+
sys.path.append(_REPO_ROOT)
|
| 49 |
+
from declip_qwenvit.model.qwen3vit_qk import block_merge_to_row_major_permutation # noqa: E402
|
| 50 |
+
|
| 51 |
+
|
| 52 |
+
class LlavaQwen3_5ViTConfig(LlavaOnevisionConfig):
|
| 53 |
+
"""LlavaOnevisionConfig 子类，vision_config 类型换成 Qwen3_5VisionConfig。
|
| 54 |
+
|
| 55 |
+
与 Qwen3-VL ViT 版的差异：
|
| 56 |
+
- sub_configs.vision_config 用 Qwen3_5VisionConfig
|
| 57 |
+
- 不再设置 deepstack_visual_indexes（Qwen3_5VisionConfig 用 AttributeError 哨兵屏蔽此字段）
|
| 58 |
+
|
| 59 |
+
其余同 LlavaQwen3VLViTConfig（tile_size 默认 384；vision_feature_select_strategy='full'
|
| 60 |
+
必须 override，Qwen3.5 ViT 无 CLS token）。
|
| 61 |
+
"""
|
| 62 |
+
|
| 63 |
+
model_type = "llava_qwen3_5vit_qwen3"
|
| 64 |
+
sub_configs = {"vision_config": Qwen3_5VisionConfig, "text_config": Qwen3Config}
|
| 65 |
+
|
| 66 |
+
def __init__(
|
| 67 |
+
self,
|
| 68 |
+
vision_config=None,
|
| 69 |
+
text_config=None,
|
| 70 |
+
tile_size: int = 384,
|
| 71 |
+
**kwargs,
|
| 72 |
+
):
|
| 73 |
+
if isinstance(vision_config, dict):
|
| 74 |
+
vision_config = Qwen3_5VisionConfig(**vision_config)
|
| 75 |
+
elif vision_config is None:
|
| 76 |
+
vision_config = Qwen3_5VisionConfig()
|
| 77 |
+
# WHY 无 `vision_config.deepstack_visual_indexes = []`（对比 Qwen3-VL ViT 版）：
|
| 78 |
+
# Qwen3_5VisionConfig 父类用 AttributeError() 哨兵显式屏蔽此字段，设置会报 AttributeError
|
| 79 |
+
# LlavaOnevision.pack_image_features 用这个作为 tile 像素大小（不是 patch_size）
|
| 80 |
+
vision_config.image_size = tile_size
|
| 81 |
+
|
| 82 |
+
if isinstance(text_config, dict):
|
| 83 |
+
text_config = Qwen3Config(**text_config)
|
| 84 |
+
elif text_config is None:
|
| 85 |
+
text_config = Qwen3Config()
|
| 86 |
+
|
| 87 |
+
# 父类默认 select_strategy='default' 会跳首 token (CLS) — Qwen3.5 ViT 无 CLS 必须用 'full'
|
| 88 |
+
kwargs.setdefault("vision_feature_select_strategy", "full")
|
| 89 |
+
super().__init__(vision_config=vision_config, text_config=text_config, **kwargs)
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
class Qwen3_5ViTBackbone(Qwen3_5VisionModel):
|
| 93 |
+
"""Qwen3.5 Vision 去除原生 patch_merger 的 backbone 版本（V6 final_layernorm fix (2026-05-16): append final_layernorm）。
|
| 94 |
+
|
| 95 |
+
构造时把 merger.norm 的预训练权重抠到 final_layernorm，然后 `del self.merger`
|
| 96 |
+
释放 ~37M 参数（保留 norm 的 LN 焊到末端做 post_layernorm 角色，丢弃 spatial
|
| 97 |
+
shuffle + linear_fc1/fc2，那对应 LlavaOV projector 的职责）。
|
| 98 |
+
|
| 99 |
+
架构对称（V6 final_layernorm fix (2026-05-16) 修复）：
|
| 100 |
+
SigLIP2: encoder → post_layernorm → last_hidden_state → LlavaOV MLP → LLM
|
| 101 |
+
V6 final_layernorm fix (2026-05-16): encoder → final_layernorm → last_hidden_state → LlavaOV MLP → LLM
|
| 102 |
+
|
| 103 |
+
forward 跑完 transformer blocks 后过 final_layernorm，再返回。下游 LlavaOnevision
|
| 104 |
+
pack_image_features 的 AnyRes 2×2 pool 接管原 merger 的空间合并职责。
|
| 105 |
+
|
| 106 |
+
输入输出契约与父类 Qwen3_5VisionModel 一致（NaViT flat）：
|
| 107 |
+
forward(hidden_states=[L, patch_dim], grid_thw=[N, 3])
|
| 108 |
+
→ BaseModelOutput(last_hidden_state=[L, hidden_size])
|
| 109 |
+
|
| 110 |
+
forward 主体 1:1 对照 `Qwen3_5VisionModel.forward`（已无 deepstack loop，比
|
| 111 |
+
Qwen3VLVisionModel.forward 更短），仅跳过末尾 `self.merger(x)`，改为 final_layernorm。
|
| 112 |
+
"""
|
| 113 |
+
|
| 114 |
+
def __init__(self, config):
|
| 115 |
+
super().__init__(config)
|
| 116 |
+
# V6 final_layernorm fix (2026-05-16): extract merger.norm pretrained weights into final_layernorm.
|
| 117 |
+
# Default init: even if ckpt 阶段没 inject final_layernorm.* (e.g. stock
|
| 118 |
+
# bootstrap path that filters merger.*), final_layernorm 仍持有 merger.norm
|
| 119 |
+
# 的预训练值, 不是 random — 这是防止 silent corruption 的兜底.
|
| 120 |
+
ln_w = self.merger.norm.weight.detach().clone()
|
| 121 |
+
ln_b = self.merger.norm.bias.detach().clone()
|
| 122 |
+
del self.merger
|
| 123 |
+
self.final_layernorm = nn.LayerNorm(config.hidden_size, eps=1e-6)
|
| 124 |
+
self.final_layernorm.weight.data.copy_(ln_w)
|
| 125 |
+
self.final_layernorm.bias.data.copy_(ln_b)
|
| 126 |
+
|
| 127 |
+
def forward(self, hidden_states, grid_thw, **kwargs):
|
| 128 |
+
hidden_states = self.patch_embed(hidden_states)
|
| 129 |
+
|
| 130 |
+
pos_embeds = self.fast_pos_embed_interpolate(grid_thw)
|
| 131 |
+
hidden_states = hidden_states + pos_embeds
|
| 132 |
+
|
| 133 |
+
rotary_pos_emb = self.rot_pos_emb(grid_thw)
|
| 134 |
+
seq_len, _ = hidden_states.size()
|
| 135 |
+
rotary_pos_emb = rotary_pos_emb.reshape(seq_len, -1)
|
| 136 |
+
emb = torch.cat((rotary_pos_emb, rotary_pos_emb), dim=-1)
|
| 137 |
+
position_embeddings = (emb.cos(), emb.sin())
|
| 138 |
+
|
| 139 |
+
cu_seqlens = torch.repeat_interleave(
|
| 140 |
+
grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]
|
| 141 |
+
).cumsum(dim=0, dtype=torch.int32)
|
| 142 |
+
cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
|
| 143 |
+
|
| 144 |
+
for blk in self.blocks:
|
| 145 |
+
hidden_states = blk(
|
| 146 |
+
hidden_states,
|
| 147 |
+
cu_seqlens=cu_seqlens,
|
| 148 |
+
position_embeddings=position_embeddings,
|
| 149 |
+
**kwargs,
|
| 150 |
+
)
|
| 151 |
+
|
| 152 |
+
# V6 final_layernorm fix (2026-05-16): appended final LayerNorm — mirrors SigLIP2's post_layernorm.
|
| 153 |
+
# Per-token affine; layout-invariant (reorder happens in adapter).
|
| 154 |
+
hidden_states = self.final_layernorm(hidden_states)
|
| 155 |
+
|
| 156 |
+
return BaseModelOutput(last_hidden_state=hidden_states)
|
| 157 |
+
|
| 158 |
+
|
| 159 |
+
class Qwen3_5ViTAsSiglipAdapter(nn.Module):
|
| 160 |
+
"""SigLIP 契约 → NaViT 契约的翻译层。持有 Qwen3_5ViTBackbone。
|
| 161 |
+
|
| 162 |
+
对外暴露 SigLIP 式 forward(pixel_values=[N,3,H,W]) → BaseModelOutputWithPooling，
|
| 163 |
+
供 LlavaOnevision 消费；对内按官方 _preprocess 的 reshape 链把 pixel_values
|
| 164 |
+
转成 NaViT flat + grid_thw 喂给 Backbone。
|
| 165 |
+
|
| 166 |
+
reshape 链 1:1 照抄 transformers 官方 Qwen2VLImageProcessorFast._preprocess
|
| 167 |
+
(video_processing_qwen3_vl.py L227-252) —— Qwen3.5 无独立 image_processor，复用 Qwen3-VL 格式。
|
| 168 |
+
"""
|
| 169 |
+
|
| 170 |
+
def __init__(self, vision_config: Qwen3_5VisionConfig):
|
| 171 |
+
super().__init__()
|
| 172 |
+
self.vision = Qwen3_5ViTBackbone(vision_config)
|
| 173 |
+
self.config = vision_config
|
| 174 |
+
|
| 175 |
+
@property
|
| 176 |
+
def dtype(self):
|
| 177 |
+
return next(self.parameters()).dtype
|
| 178 |
+
|
| 179 |
+
@property
|
| 180 |
+
def device(self):
|
| 181 |
+
return next(self.parameters()).device
|
| 182 |
+
|
| 183 |
+
def _flatten_navit(self, pixel_values: torch.Tensor):
|
| 184 |
+
"""[N, 3, H, W] → (flat=[N*L, patch_dim], grid_thw=[N, 3], shape=(N, L)).
|
| 185 |
+
|
| 186 |
+
L = grid_t * grid_h * grid_w = 1 * (H/16) * (W/16)
|
| 187 |
+
patch_dim = C * temporal_patch_size * patch_size^2 = 3 * 2 * 16 * 16 = 1536
|
| 188 |
+
"""
|
| 189 |
+
pixel_values = pixel_values.to(dtype=self.dtype)
|
| 190 |
+
tps = self.config.temporal_patch_size
|
| 191 |
+
ps = self.config.patch_size
|
| 192 |
+
ms = self.config.spatial_merge_size
|
| 193 |
+
|
| 194 |
+
patches = pixel_values.unsqueeze(1)
|
| 195 |
+
# 对单帧图像 T=1, pad=1 → expand 一帧使 T 整除 temporal_patch_size，
|
| 196 |
+
# Conv3d 在复制帧上退化为等效 2D Conv（数学无损）
|
| 197 |
+
T = patches.shape[1]
|
| 198 |
+
pad = -T % tps
|
| 199 |
+
if pad:
|
| 200 |
+
repeats = patches[:, -1:].expand(-1, pad, -1, -1, -1)
|
| 201 |
+
patches = torch.cat((patches, repeats), dim=1)
|
| 202 |
+
|
| 203 |
+
batch_size, t, channel, H, W = patches.shape
|
| 204 |
+
grid_t = t // tps
|
| 205 |
+
grid_h = H // ps
|
| 206 |
+
grid_w = W // ps
|
| 207 |
+
|
| 208 |
+
patches = patches.view(
|
| 209 |
+
batch_size, grid_t, tps, channel,
|
| 210 |
+
grid_h // ms, ms, ps,
|
| 211 |
+
grid_w // ms, ms, ps,
|
| 212 |
+
)
|
| 213 |
+
patches = patches.permute(0, 1, 4, 7, 5, 8, 3, 2, 6, 9)
|
| 214 |
+
flatten_patches = patches.reshape(
|
| 215 |
+
batch_size,
|
| 216 |
+
grid_t * grid_h * grid_w,
|
| 217 |
+
channel * tps * ps * ps,
|
| 218 |
+
)
|
| 219 |
+
|
| 220 |
+
seq_len = grid_t * grid_h * grid_w
|
| 221 |
+
flat = flatten_patches.reshape(batch_size * seq_len, -1)
|
| 222 |
+
# on-device 构造小 tensor 再 expand，host→GPU 同步量 O(3) 而非 O(N*3)
|
| 223 |
+
grid_unit = torch.tensor(
|
| 224 |
+
[grid_t, grid_h, grid_w], dtype=torch.int32, device=pixel_values.device,
|
| 225 |
+
)
|
| 226 |
+
grid_thw = grid_unit.unsqueeze(0).expand(batch_size, -1).contiguous()
|
| 227 |
+
return flat, grid_thw, (batch_size, seq_len)
|
| 228 |
+
|
| 229 |
+
def forward(
|
| 230 |
+
self,
|
| 231 |
+
pixel_values: torch.Tensor,
|
| 232 |
+
output_hidden_states: Optional[bool] = None,
|
| 233 |
+
return_dict: Optional[bool] = None,
|
| 234 |
+
**kwargs,
|
| 235 |
+
) -> BaseModelOutputWithPooling:
|
| 236 |
+
flat, grid_thw, (N, S) = self._flatten_navit(pixel_values)
|
| 237 |
+
vision_out = self.vision(flat, grid_thw=grid_thw)
|
| 238 |
+
hidden = vision_out.last_hidden_state.view(N, S, -1)
|
| 239 |
+
|
| 240 |
+
# Block-merge → row-major reorder before handing to LlavaOnevision.
|
| 241 |
+
# Internally the ViT runs in Qwen NaViT block-merge layout (pretrained
|
| 242 |
+
# pos_embed + RoPE contract); downstream LlavaOV `pack_image_features`
|
| 243 |
+
# (multi-tile AnyRes path, view(num_patch_h, num_patch_w, h, w, -1))
|
| 244 |
+
# and `apply_pooling` (video path, view(B, h, w, -1) + bilinear) BOTH
|
| 245 |
+
# assume row-major. Without this reorder, the multi-tile/video spatial
|
| 246 |
+
# pool pulls together tokens that are NOT spatially adjacent — silent
|
| 247 |
+
# corruption that doesn't fire on S1 single-tile path (line 348-351 of
|
| 248 |
+
# modeling_llava_onevision.py just flattens [N,D] verbatim) but kills
|
| 249 |
+
# S2 / eval quality.
|
| 250 |
+
grid_h = int(grid_thw[0, 1].item())
|
| 251 |
+
grid_w = int(grid_thw[0, 2].item())
|
| 252 |
+
ms = getattr(self.config, "spatial_merge_size", 2)
|
| 253 |
+
layout_perm = block_merge_to_row_major_permutation(
|
| 254 |
+
grid_h, grid_w, ms=ms, device=hidden.device,
|
| 255 |
+
)
|
| 256 |
+
hidden = hidden[:, layout_perm, :]
|
| 257 |
+
|
| 258 |
+
return BaseModelOutputWithPooling(
|
| 259 |
+
last_hidden_state=hidden,
|
| 260 |
+
# LlavaOnevision 索引 hidden_states[vision_feature_layer=-1]；tuple 长度 1 足够
|
| 261 |
+
hidden_states=(hidden,),
|
| 262 |
+
pooler_output=None,
|
| 263 |
+
)
|
| 264 |
+
|
| 265 |
+
|
| 266 |
+
class LlavaQwen3_5ViTMultiModalProjector(nn.Module):
|
| 267 |
+
"""标准 LlavaOnevision projector（V6 final_layernorm fix (2026-05-16): pre_norm → Identity）。
|
| 268 |
+
|
| 269 |
+
V6 final_layernorm fix (2026-05-16) 修复后, encoder 末端已自带 final_layernorm（与 SigLIP2 post_layernorm 对称），
|
| 270 |
+
projector 不再需要补 LN — pre_norm 改为 nn.Identity，对齐 SigLIP2 plugin 的
|
| 271 |
+
LlavaOnevision stock projector 结构（裸 linear_1 → GELU → linear_2），
|
| 272 |
+
保证 SigLIP2 / Qwen3.5 / Qwen3-VL 三个 backbone 在 LlavaOV 设定下公平对比。
|
| 273 |
+
|
| 274 |
+
（历史：V6.0.0~V6.0.4 时期 encoder 无 final LN，projector pre_norm 是补丁；
|
| 275 |
+
现在补丁回到 encoder 内部，projector 回归 stock 形态。）
|
| 276 |
+
"""
|
| 277 |
+
|
| 278 |
+
def __init__(self, config: LlavaQwen3_5ViTConfig):
|
| 279 |
+
super().__init__()
|
| 280 |
+
num_feature_layers = (
|
| 281 |
+
1 if isinstance(config.vision_feature_layer, int) else len(config.vision_feature_layer)
|
| 282 |
+
)
|
| 283 |
+
vision_dim = config.vision_config.hidden_size * num_feature_layers
|
| 284 |
+
text_dim = config.text_config.hidden_size
|
| 285 |
+
bias = getattr(config, "multimodal_projector_bias", True)
|
| 286 |
+
|
| 287 |
+
# V6 final_layernorm fix (2026-05-16): pre_norm = Identity (encoder 已自带 final_layernorm).
|
| 288 |
+
self.pre_norm = nn.Identity()
|
| 289 |
+
self.linear_1 = nn.Linear(vision_dim, text_dim, bias=bias)
|
| 290 |
+
self.act = ACT2FN[config.projector_hidden_act]
|
| 291 |
+
self.linear_2 = nn.Linear(text_dim, text_dim, bias=bias)
|
| 292 |
+
|
| 293 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
| 294 |
+
return self.linear_2(self.act(self.linear_1(self.pre_norm(x))))
|
| 295 |
+
|
| 296 |
+
|
| 297 |
+
class LlavaQwen3_5ViTModel(LlavaOnevisionModel):
|
| 298 |
+
"""继承 LlavaOnevisionModel 但绕过其 __init__ 手动装配。
|
| 299 |
+
|
| 300 |
+
父类 __init__ 调 `AutoModel.from_config(config.vision_config)` 会对 Qwen3_5VisionConfig
|
| 301 |
+
抛 "Unrecognized configuration"（Qwen3.5 vision 没注册到 AutoModel）。手动装配避开
|
| 302 |
+
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）。
|
| 303 |
+
|
| 304 |
+
装配顺序与父类一致：vision_tower / projector / image_newline / language_model / post_init。
|
| 305 |
+
"""
|
| 306 |
+
|
| 307 |
+
config_class = LlavaQwen3_5ViTConfig
|
| 308 |
+
|
| 309 |
+
def __init__(self, config: LlavaQwen3_5ViTConfig):
|
| 310 |
+
# 跳过 LlavaOnevisionModel.__init__（AutoModel 不识别 Qwen3_5VisionConfig）
|
| 311 |
+
LlavaOnevisionPreTrainedModel.__init__(self, config)
|
| 312 |
+
self.vision_tower = Qwen3_5ViTAsSiglipAdapter(config.vision_config)
|
| 313 |
+
self.multi_modal_projector = LlavaQwen3_5ViTMultiModalProjector(config)
|
| 314 |
+
embed_std = 1 / math.sqrt(config.text_config.hidden_size)
|
| 315 |
+
self.image_newline = nn.Parameter(
|
| 316 |
+
torch.randn(config.text_config.hidden_size, dtype=self.dtype) * embed_std
|
| 317 |
+
)
|
| 318 |
+
self.vocab_size = config.text_config.vocab_size
|
| 319 |
+
self.language_model = AutoModel.from_config(config.text_config)
|
| 320 |
+
self.post_init()
|
| 321 |
+
|
| 322 |
+
|
| 323 |
+
class LlavaQwen3_5ViTForConditionalGeneration(LlavaOnevisionForConditionalGeneration):
|
| 324 |
+
"""继承 LlavaOnevisionForConditionalGeneration，只换 self.model。
|
| 325 |
+
|
| 326 |
+
同样跳过父类 __init__（避免重复构造 LlavaOnevisionModel，根因见 LlavaQwen3_5ViTModel）。
|
| 327 |
+
"""
|
| 328 |
+
|
| 329 |
+
config_class = LlavaQwen3_5ViTConfig
|
| 330 |
+
|
| 331 |
+
def __init__(self, config: LlavaQwen3_5ViTConfig):
|
| 332 |
+
LlavaOnevisionPreTrainedModel.__init__(self, config)
|
| 333 |
+
self.model = LlavaQwen3_5ViTModel(config)
|
| 334 |
+
self.lm_head = nn.Linear(
|
| 335 |
+
config.text_config.hidden_size, config.text_config.vocab_size, bias=False
|
| 336 |
+
)
|
| 337 |
+
self.post_init()
|
| 338 |
+
|
| 339 |
+
|
| 340 |
+
AutoConfig.register(LlavaQwen3_5ViTConfig.model_type, LlavaQwen3_5ViTConfig)
|
| 341 |
+
AutoModelForCausalLM.register(LlavaQwen3_5ViTConfig, LlavaQwen3_5ViTForConditionalGeneration)
|
| 342 |
+
|
| 343 |
+
|
| 344 |
+
__all__ = [
|
| 345 |
+
"LlavaQwen3_5ViTConfig",
|
| 346 |
+
"Qwen3_5ViTBackbone",
|
| 347 |
+
"Qwen3_5ViTAsSiglipAdapter",
|
| 348 |
+
"LlavaQwen3_5ViTMultiModalProjector",
|
| 349 |
+
"LlavaQwen3_5ViTModel",
|
| 350 |
+
"LlavaQwen3_5ViTForConditionalGeneration",
|
| 351 |
+
]
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/processor_config.json
ADDED
|
@@ -0,0 +1,203 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"image_processor": {
|
| 3 |
+
"do_convert_rgb": true,
|
| 4 |
+
"do_normalize": true,
|
| 5 |
+
"do_pad": true,
|
| 6 |
+
"do_rescale": true,
|
| 7 |
+
"do_resize": true,
|
| 8 |
+
"image_grid_pinpoints": [
|
| 9 |
+
[
|
| 10 |
+
384,
|
| 11 |
+
384
|
| 12 |
+
],
|
| 13 |
+
[
|
| 14 |
+
384,
|
| 15 |
+
768
|
| 16 |
+
],
|
| 17 |
+
[
|
| 18 |
+
384,
|
| 19 |
+
1152
|
| 20 |
+
],
|
| 21 |
+
[
|
| 22 |
+
384,
|
| 23 |
+
1536
|
| 24 |
+
],
|
| 25 |
+
[
|
| 26 |
+
384,
|
| 27 |
+
1920
|
| 28 |
+
],
|
| 29 |
+
[
|
| 30 |
+
384,
|
| 31 |
+
2304
|
| 32 |
+
],
|
| 33 |
+
[
|
| 34 |
+
768,
|
| 35 |
+
384
|
| 36 |
+
],
|
| 37 |
+
[
|
| 38 |
+
768,
|
| 39 |
+
768
|
| 40 |
+
],
|
| 41 |
+
[
|
| 42 |
+
768,
|
| 43 |
+
1152
|
| 44 |
+
],
|
| 45 |
+
[
|
| 46 |
+
768,
|
| 47 |
+
1536
|
| 48 |
+
],
|
| 49 |
+
[
|
| 50 |
+
768,
|
| 51 |
+
1920
|
| 52 |
+
],
|
| 53 |
+
[
|
| 54 |
+
768,
|
| 55 |
+
2304
|
| 56 |
+
],
|
| 57 |
+
[
|
| 58 |
+
1152,
|
| 59 |
+
384
|
| 60 |
+
],
|
| 61 |
+
[
|
| 62 |
+
1152,
|
| 63 |
+
768
|
| 64 |
+
],
|
| 65 |
+
[
|
| 66 |
+
1152,
|
| 67 |
+
1152
|
| 68 |
+
],
|
| 69 |
+
[
|
| 70 |
+
1152,
|
| 71 |
+
1536
|
| 72 |
+
],
|
| 73 |
+
[
|
| 74 |
+
1152,
|
| 75 |
+
1920
|
| 76 |
+
],
|
| 77 |
+
[
|
| 78 |
+
1152,
|
| 79 |
+
2304
|
| 80 |
+
],
|
| 81 |
+
[
|
| 82 |
+
1536,
|
| 83 |
+
384
|
| 84 |
+
],
|
| 85 |
+
[
|
| 86 |
+
1536,
|
| 87 |
+
768
|
| 88 |
+
],
|
| 89 |
+
[
|
| 90 |
+
1536,
|
| 91 |
+
1152
|
| 92 |
+
],
|
| 93 |
+
[
|
| 94 |
+
1536,
|
| 95 |
+
1536
|
| 96 |
+
],
|
| 97 |
+
[
|
| 98 |
+
1536,
|
| 99 |
+
1920
|
| 100 |
+
],
|
| 101 |
+
[
|
| 102 |
+
1536,
|
| 103 |
+
2304
|
| 104 |
+
],
|
| 105 |
+
[
|
| 106 |
+
1920,
|
| 107 |
+
384
|
| 108 |
+
],
|
| 109 |
+
[
|
| 110 |
+
1920,
|
| 111 |
+
768
|
| 112 |
+
],
|
| 113 |
+
[
|
| 114 |
+
1920,
|
| 115 |
+
1152
|
| 116 |
+
],
|
| 117 |
+
[
|
| 118 |
+
1920,
|
| 119 |
+
1536
|
| 120 |
+
],
|
| 121 |
+
[
|
| 122 |
+
1920,
|
| 123 |
+
1920
|
| 124 |
+
],
|
| 125 |
+
[
|
| 126 |
+
1920,
|
| 127 |
+
2304
|
| 128 |
+
],
|
| 129 |
+
[
|
| 130 |
+
2304,
|
| 131 |
+
384
|
| 132 |
+
],
|
| 133 |
+
[
|
| 134 |
+
2304,
|
| 135 |
+
768
|
| 136 |
+
],
|
| 137 |
+
[
|
| 138 |
+
2304,
|
| 139 |
+
1152
|
| 140 |
+
],
|
| 141 |
+
[
|
| 142 |
+
2304,
|
| 143 |
+
1536
|
| 144 |
+
],
|
| 145 |
+
[
|
| 146 |
+
2304,
|
| 147 |
+
1920
|
| 148 |
+
],
|
| 149 |
+
[
|
| 150 |
+
2304,
|
| 151 |
+
2304
|
| 152 |
+
]
|
| 153 |
+
],
|
| 154 |
+
"image_mean": [
|
| 155 |
+
0.5,
|
| 156 |
+
0.5,
|
| 157 |
+
0.5
|
| 158 |
+
],
|
| 159 |
+
"image_processor_type": "LlavaOnevisionImageProcessor",
|
| 160 |
+
"image_std": [
|
| 161 |
+
0.5,
|
| 162 |
+
0.5,
|
| 163 |
+
0.5
|
| 164 |
+
],
|
| 165 |
+
"resample": 3,
|
| 166 |
+
"rescale_factor": 0.00392156862745098,
|
| 167 |
+
"size": {
|
| 168 |
+
"height": 384,
|
| 169 |
+
"width": 384
|
| 170 |
+
}
|
| 171 |
+
},
|
| 172 |
+
"image_token": "<image>",
|
| 173 |
+
"num_image_tokens": 576,
|
| 174 |
+
"processor_class": "LlavaOnevisionProcessor",
|
| 175 |
+
"video_processor": {
|
| 176 |
+
"do_convert_rgb": true,
|
| 177 |
+
"do_normalize": true,
|
| 178 |
+
"do_rescale": true,
|
| 179 |
+
"do_resize": true,
|
| 180 |
+
"do_sample_frames": false,
|
| 181 |
+
"image_mean": [
|
| 182 |
+
0.5,
|
| 183 |
+
0.5,
|
| 184 |
+
0.5
|
| 185 |
+
],
|
| 186 |
+
"image_std": [
|
| 187 |
+
0.5,
|
| 188 |
+
0.5,
|
| 189 |
+
0.5
|
| 190 |
+
],
|
| 191 |
+
"resample": 3,
|
| 192 |
+
"rescale_factor": 0.00392156862745098,
|
| 193 |
+
"return_metadata": false,
|
| 194 |
+
"size": {
|
| 195 |
+
"height": 384,
|
| 196 |
+
"width": 384
|
| 197 |
+
},
|
| 198 |
+
"video_processor_type": "LlavaOnevisionVideoProcessor"
|
| 199 |
+
},
|
| 200 |
+
"video_token": "<video>",
|
| 201 |
+
"vision_aspect_ratio": "anyres_max_9",
|
| 202 |
+
"vision_feature_select_strategy": null
|
| 203 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9024318c850eaadf26be79389d21b07a7afd8f1af749b89f9109b06c0d12173c
|
| 3 |
+
size 11423018
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/tokenizer_config.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": null,
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|im_end|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<image>",
|
| 10 |
+
"<video>"
|
| 11 |
+
],
|
| 12 |
+
"is_local": true,
|
| 13 |
+
"model_max_length": 1010000,
|
| 14 |
+
"pad_token": "<|endoftext|>",
|
| 15 |
+
"processor_class": "LlavaOnevisionProcessor",
|
| 16 |
+
"split_special_tokens": false,
|
| 17 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 18 |
+
"unk_token": null
|
| 19 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/stock_16f_10pct/v0-20260530-001435/checkpoint-505/zero_to_fp32.py
ADDED
|
@@ -0,0 +1,760 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python
|
| 2 |
+
|
| 3 |
+
# Copyright (c) Microsoft Corporation.
|
| 4 |
+
# SPDX-License-Identifier: Apache-2.0
|
| 5 |
+
|
| 6 |
+
# DeepSpeed Team
|
| 7 |
+
|
| 8 |
+
# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
|
| 9 |
+
# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
|
| 10 |
+
# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
|
| 11 |
+
# application.
|
| 12 |
+
#
|
| 13 |
+
# example:
|
| 14 |
+
# python zero_to_fp32.py . output_dir/
|
| 15 |
+
# or
|
| 16 |
+
# python zero_to_fp32.py . output_dir/ --safe_serialization
|
| 17 |
+
|
| 18 |
+
import argparse
|
| 19 |
+
import torch
|
| 20 |
+
import glob
|
| 21 |
+
import math
|
| 22 |
+
import os
|
| 23 |
+
import re
|
| 24 |
+
import gc
|
| 25 |
+
import json
|
| 26 |
+
import numpy as np
|
| 27 |
+
from tqdm import tqdm
|
| 28 |
+
from collections import OrderedDict
|
| 29 |
+
from dataclasses import dataclass
|
| 30 |
+
|
| 31 |
+
# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
|
| 32 |
+
# DeepSpeed data structures it has to be available in the current python environment.
|
| 33 |
+
from deepspeed.utils import logger
|
| 34 |
+
from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
|
| 35 |
+
FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
|
| 36 |
+
FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
@dataclass
|
| 40 |
+
class zero_model_state:
|
| 41 |
+
buffers: dict()
|
| 42 |
+
param_shapes: dict()
|
| 43 |
+
shared_params: list
|
| 44 |
+
ds_version: int
|
| 45 |
+
frozen_param_shapes: dict()
|
| 46 |
+
frozen_param_fragments: dict()
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
debug = 0
|
| 50 |
+
|
| 51 |
+
# load to cpu
|
| 52 |
+
device = torch.device('cpu')
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
def atoi(text):
|
| 56 |
+
return int(text) if text.isdigit() else text
|
| 57 |
+
|
| 58 |
+
|
| 59 |
+
def natural_keys(text):
|
| 60 |
+
'''
|
| 61 |
+
alist.sort(key=natural_keys) sorts in human order
|
| 62 |
+
http://nedbatchelder.com/blog/200712/human_sorting.html
|
| 63 |
+
(See Toothy's implementation in the comments)
|
| 64 |
+
'''
|
| 65 |
+
return [atoi(c) for c in re.split(r'(\d+)', text)]
|
| 66 |
+
|
| 67 |
+
|
| 68 |
+
def get_model_state_file(checkpoint_dir, zero_stage):
|
| 69 |
+
if not os.path.isdir(checkpoint_dir):
|
| 70 |
+
raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
|
| 71 |
+
|
| 72 |
+
# there should be only one file
|
| 73 |
+
if zero_stage <= 2:
|
| 74 |
+
file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
|
| 75 |
+
elif zero_stage == 3:
|
| 76 |
+
file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
|
| 77 |
+
|
| 78 |
+
if not os.path.exists(file):
|
| 79 |
+
raise FileNotFoundError(f"can't find model states file at '{file}'")
|
| 80 |
+
|
| 81 |
+
return file
|
| 82 |
+
|
| 83 |
+
|
| 84 |
+
def get_checkpoint_files(checkpoint_dir, glob_pattern):
|
| 85 |
+
# XXX: need to test that this simple glob rule works for multi-node setup too
|
| 86 |
+
ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
|
| 87 |
+
|
| 88 |
+
if len(ckpt_files) == 0:
|
| 89 |
+
raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
|
| 90 |
+
|
| 91 |
+
return ckpt_files
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
def get_optim_files(checkpoint_dir):
|
| 95 |
+
return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
|
| 96 |
+
|
| 97 |
+
|
| 98 |
+
def get_model_state_files(checkpoint_dir):
|
| 99 |
+
return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
def parse_model_states(files):
|
| 103 |
+
zero_model_states = []
|
| 104 |
+
for file in files:
|
| 105 |
+
state_dict = torch.load(file, map_location=device, weights_only=False)
|
| 106 |
+
|
| 107 |
+
if BUFFER_NAMES not in state_dict:
|
| 108 |
+
raise ValueError(f"{file} is not a model state checkpoint")
|
| 109 |
+
buffer_names = state_dict[BUFFER_NAMES]
|
| 110 |
+
if debug:
|
| 111 |
+
print("Found buffers:", buffer_names)
|
| 112 |
+
|
| 113 |
+
# recover just the buffers while restoring them to fp32 if they were saved in fp16
|
| 114 |
+
buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
|
| 115 |
+
param_shapes = state_dict[PARAM_SHAPES]
|
| 116 |
+
|
| 117 |
+
# collect parameters that are included in param_shapes
|
| 118 |
+
param_names = []
|
| 119 |
+
for s in param_shapes:
|
| 120 |
+
for name in s.keys():
|
| 121 |
+
param_names.append(name)
|
| 122 |
+
|
| 123 |
+
# update with frozen parameters
|
| 124 |
+
frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
|
| 125 |
+
if frozen_param_shapes is not None:
|
| 126 |
+
if debug:
|
| 127 |
+
print(f"Found frozen_param_shapes: {frozen_param_shapes}")
|
| 128 |
+
param_names += list(frozen_param_shapes.keys())
|
| 129 |
+
|
| 130 |
+
# handle shared params
|
| 131 |
+
shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
|
| 132 |
+
|
| 133 |
+
ds_version = state_dict.get(DS_VERSION, None)
|
| 134 |
+
|
| 135 |
+
frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
|
| 136 |
+
|
| 137 |
+
z_model_state = zero_model_state(buffers=buffers,
|
| 138 |
+
param_shapes=param_shapes,
|
| 139 |
+
shared_params=shared_params,
|
| 140 |
+
ds_version=ds_version,
|
| 141 |
+
frozen_param_shapes=frozen_param_shapes,
|
| 142 |
+
frozen_param_fragments=frozen_param_fragments)
|
| 143 |
+
zero_model_states.append(z_model_state)
|
| 144 |
+
|
| 145 |
+
return zero_model_states
|
| 146 |
+
|
| 147 |
+
|
| 148 |
+
def parse_optim_states(files, ds_checkpoint_dir):
|
| 149 |
+
total_files = len(files)
|
| 150 |
+
state_dicts = []
|
| 151 |
+
for f in tqdm(files, desc='Loading checkpoint shards'):
|
| 152 |
+
state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
|
| 153 |
+
# immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
|
| 154 |
+
# and also handle the case where it was already removed by another helper script
|
| 155 |
+
state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
|
| 156 |
+
state_dicts.append(state_dict)
|
| 157 |
+
|
| 158 |
+
if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
|
| 159 |
+
raise ValueError(f"{files[0]} is not a zero checkpoint")
|
| 160 |
+
zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
|
| 161 |
+
world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
|
| 162 |
+
|
| 163 |
+
# For ZeRO-2 each param group can have different partition_count as data parallelism for expert
|
| 164 |
+
# parameters can be different from data parallelism for non-expert parameters. So we can just
|
| 165 |
+
# use the max of the partition_count to get the dp world_size.
|
| 166 |
+
|
| 167 |
+
if type(world_size) is list:
|
| 168 |
+
world_size = max(world_size)
|
| 169 |
+
|
| 170 |
+
if world_size != total_files:
|
| 171 |
+
raise ValueError(
|
| 172 |
+
f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
|
| 173 |
+
"Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
|
| 174 |
+
)
|
| 175 |
+
|
| 176 |
+
# the groups are named differently in each stage
|
| 177 |
+
if zero_stage <= 2:
|
| 178 |
+
fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
|
| 179 |
+
elif zero_stage == 3:
|
| 180 |
+
fp32_groups_key = FP32_FLAT_GROUPS
|
| 181 |
+
else:
|
| 182 |
+
raise ValueError(f"unknown zero stage {zero_stage}")
|
| 183 |
+
|
| 184 |
+
fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
|
| 185 |
+
return zero_stage, world_size, fp32_flat_groups
|
| 186 |
+
|
| 187 |
+
|
| 188 |
+
def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
|
| 189 |
+
"""
|
| 190 |
+
Returns fp32 state_dict reconstructed from ds checkpoint
|
| 191 |
+
|
| 192 |
+
Args:
|
| 193 |
+
- ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
|
| 194 |
+
|
| 195 |
+
"""
|
| 196 |
+
print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
|
| 197 |
+
|
| 198 |
+
optim_files = get_optim_files(ds_checkpoint_dir)
|
| 199 |
+
zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
|
| 200 |
+
print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
|
| 201 |
+
|
| 202 |
+
model_files = get_model_state_files(ds_checkpoint_dir)
|
| 203 |
+
|
| 204 |
+
zero_model_states = parse_model_states(model_files)
|
| 205 |
+
print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
|
| 206 |
+
|
| 207 |
+
if zero_stage <= 2:
|
| 208 |
+
return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
| 209 |
+
exclude_frozen_parameters)
|
| 210 |
+
elif zero_stage == 3:
|
| 211 |
+
return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
| 212 |
+
exclude_frozen_parameters)
|
| 213 |
+
|
| 214 |
+
|
| 215 |
+
def _zero2_merge_frozen_params(state_dict, zero_model_states):
|
| 216 |
+
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
| 217 |
+
return
|
| 218 |
+
|
| 219 |
+
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
| 220 |
+
frozen_param_fragments = zero_model_states[0].frozen_param_fragments
|
| 221 |
+
|
| 222 |
+
if debug:
|
| 223 |
+
num_elem = sum(s.numel() for s in frozen_param_shapes.values())
|
| 224 |
+
print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
| 225 |
+
|
| 226 |
+
wanted_params = len(frozen_param_shapes)
|
| 227 |
+
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
| 228 |
+
avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
|
| 229 |
+
print(f'Frozen params: Have {avail_numel} numels to process.')
|
| 230 |
+
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
| 231 |
+
|
| 232 |
+
total_params = 0
|
| 233 |
+
total_numel = 0
|
| 234 |
+
for name, shape in frozen_param_shapes.items():
|
| 235 |
+
total_params += 1
|
| 236 |
+
unpartitioned_numel = shape.numel()
|
| 237 |
+
total_numel += unpartitioned_numel
|
| 238 |
+
|
| 239 |
+
state_dict[name] = frozen_param_fragments[name]
|
| 240 |
+
|
| 241 |
+
if debug:
|
| 242 |
+
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
| 243 |
+
|
| 244 |
+
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
| 245 |
+
|
| 246 |
+
|
| 247 |
+
def _has_callable(obj, fn):
|
| 248 |
+
attr = getattr(obj, fn, None)
|
| 249 |
+
return callable(attr)
|
| 250 |
+
|
| 251 |
+
|
| 252 |
+
def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
| 253 |
+
param_shapes = zero_model_states[0].param_shapes
|
| 254 |
+
|
| 255 |
+
# Reconstruction protocol:
|
| 256 |
+
#
|
| 257 |
+
# XXX: document this
|
| 258 |
+
|
| 259 |
+
if debug:
|
| 260 |
+
for i in range(world_size):
|
| 261 |
+
for j in range(len(fp32_flat_groups[0])):
|
| 262 |
+
print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
|
| 263 |
+
|
| 264 |
+
# XXX: memory usage doubles here (zero2)
|
| 265 |
+
num_param_groups = len(fp32_flat_groups[0])
|
| 266 |
+
merged_single_partition_of_fp32_groups = []
|
| 267 |
+
for i in range(num_param_groups):
|
| 268 |
+
merged_partitions = [sd[i] for sd in fp32_flat_groups]
|
| 269 |
+
full_single_fp32_vector = torch.cat(merged_partitions, 0)
|
| 270 |
+
merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
|
| 271 |
+
avail_numel = sum(
|
| 272 |
+
[full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
|
| 273 |
+
|
| 274 |
+
if debug:
|
| 275 |
+
wanted_params = sum([len(shapes) for shapes in param_shapes])
|
| 276 |
+
wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
|
| 277 |
+
# not asserting if there is a mismatch due to possible padding
|
| 278 |
+
print(f"Have {avail_numel} numels to process.")
|
| 279 |
+
print(f"Need {wanted_numel} numels in {wanted_params} params.")
|
| 280 |
+
|
| 281 |
+
# params
|
| 282 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
| 283 |
+
# out-of-core computing solution
|
| 284 |
+
total_numel = 0
|
| 285 |
+
total_params = 0
|
| 286 |
+
for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
|
| 287 |
+
offset = 0
|
| 288 |
+
avail_numel = full_single_fp32_vector.numel()
|
| 289 |
+
for name, shape in shapes.items():
|
| 290 |
+
|
| 291 |
+
unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
|
| 292 |
+
total_numel += unpartitioned_numel
|
| 293 |
+
total_params += 1
|
| 294 |
+
|
| 295 |
+
if debug:
|
| 296 |
+
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
| 297 |
+
state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
|
| 298 |
+
offset += unpartitioned_numel
|
| 299 |
+
|
| 300 |
+
# Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
|
| 301 |
+
# avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
|
| 302 |
+
# paddings performed in the code it's almost impossible to predict the exact numbers w/o the
|
| 303 |
+
# live optimizer object, so we are checking that the numbers are within the right range
|
| 304 |
+
align_to = 2 * world_size
|
| 305 |
+
|
| 306 |
+
def zero2_align(x):
|
| 307 |
+
return align_to * math.ceil(x / align_to)
|
| 308 |
+
|
| 309 |
+
if debug:
|
| 310 |
+
print(f"original offset={offset}, avail_numel={avail_numel}")
|
| 311 |
+
|
| 312 |
+
offset = zero2_align(offset)
|
| 313 |
+
avail_numel = zero2_align(avail_numel)
|
| 314 |
+
|
| 315 |
+
if debug:
|
| 316 |
+
print(f"aligned offset={offset}, avail_numel={avail_numel}")
|
| 317 |
+
|
| 318 |
+
# Sanity check
|
| 319 |
+
if offset != avail_numel:
|
| 320 |
+
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
| 321 |
+
|
| 322 |
+
print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
|
| 323 |
+
|
| 324 |
+
|
| 325 |
+
def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
| 326 |
+
exclude_frozen_parameters):
|
| 327 |
+
state_dict = OrderedDict()
|
| 328 |
+
|
| 329 |
+
# buffers
|
| 330 |
+
buffers = zero_model_states[0].buffers
|
| 331 |
+
state_dict.update(buffers)
|
| 332 |
+
if debug:
|
| 333 |
+
print(f"added {len(buffers)} buffers")
|
| 334 |
+
|
| 335 |
+
if not exclude_frozen_parameters:
|
| 336 |
+
_zero2_merge_frozen_params(state_dict, zero_model_states)
|
| 337 |
+
|
| 338 |
+
_zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
| 339 |
+
|
| 340 |
+
# recover shared parameters
|
| 341 |
+
for pair in zero_model_states[0].shared_params:
|
| 342 |
+
if pair[1] in state_dict:
|
| 343 |
+
state_dict[pair[0]] = state_dict[pair[1]]
|
| 344 |
+
|
| 345 |
+
return state_dict
|
| 346 |
+
|
| 347 |
+
|
| 348 |
+
def zero3_partitioned_param_info(unpartitioned_numel, world_size):
|
| 349 |
+
remainder = unpartitioned_numel % world_size
|
| 350 |
+
padding_numel = (world_size - remainder) if remainder else 0
|
| 351 |
+
partitioned_numel = math.ceil(unpartitioned_numel / world_size)
|
| 352 |
+
return partitioned_numel, padding_numel
|
| 353 |
+
|
| 354 |
+
|
| 355 |
+
def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
|
| 356 |
+
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
| 357 |
+
return
|
| 358 |
+
|
| 359 |
+
if debug:
|
| 360 |
+
for i in range(world_size):
|
| 361 |
+
num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
|
| 362 |
+
print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
| 363 |
+
|
| 364 |
+
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
| 365 |
+
wanted_params = len(frozen_param_shapes)
|
| 366 |
+
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
| 367 |
+
avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
|
| 368 |
+
print(f'Frozen params: Have {avail_numel} numels to process.')
|
| 369 |
+
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
| 370 |
+
|
| 371 |
+
total_params = 0
|
| 372 |
+
total_numel = 0
|
| 373 |
+
for name, shape in zero_model_states[0].frozen_param_shapes.items():
|
| 374 |
+
total_params += 1
|
| 375 |
+
unpartitioned_numel = shape.numel()
|
| 376 |
+
total_numel += unpartitioned_numel
|
| 377 |
+
|
| 378 |
+
param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
|
| 379 |
+
state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
|
| 380 |
+
|
| 381 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
| 382 |
+
|
| 383 |
+
if debug:
|
| 384 |
+
print(
|
| 385 |
+
f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
| 386 |
+
)
|
| 387 |
+
|
| 388 |
+
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
| 389 |
+
|
| 390 |
+
|
| 391 |
+
class GatheredTensor:
|
| 392 |
+
"""
|
| 393 |
+
A pseudo tensor that collects partitioned weights.
|
| 394 |
+
It is more memory efficient when there are multiple groups.
|
| 395 |
+
"""
|
| 396 |
+
|
| 397 |
+
def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
|
| 398 |
+
self.flat_groups = flat_groups
|
| 399 |
+
self.flat_groups_offset = flat_groups_offset
|
| 400 |
+
self.offset = offset
|
| 401 |
+
self.partitioned_numel = partitioned_numel
|
| 402 |
+
self.shape = shape
|
| 403 |
+
self.dtype = self.flat_groups[0][0].dtype
|
| 404 |
+
|
| 405 |
+
def contiguous(self):
|
| 406 |
+
"""
|
| 407 |
+
Merge partitioned weights from flat_groups into a single tensor.
|
| 408 |
+
"""
|
| 409 |
+
end_idx = self.offset + self.partitioned_numel
|
| 410 |
+
world_size = len(self.flat_groups)
|
| 411 |
+
pad_flat_param_chunks = []
|
| 412 |
+
|
| 413 |
+
for rank_i in range(world_size):
|
| 414 |
+
# for each rank, we need to collect weights from related group/groups
|
| 415 |
+
flat_groups_at_rank_i = self.flat_groups[rank_i]
|
| 416 |
+
start_group_id = None
|
| 417 |
+
end_group_id = None
|
| 418 |
+
for group_id in range(len(self.flat_groups_offset)):
|
| 419 |
+
if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
|
| 420 |
+
start_group_id = group_id
|
| 421 |
+
if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
|
| 422 |
+
end_group_id = group_id
|
| 423 |
+
break
|
| 424 |
+
# collect weights from related group/groups
|
| 425 |
+
for group_id in range(start_group_id, end_group_id + 1):
|
| 426 |
+
flat_tensor = flat_groups_at_rank_i[group_id]
|
| 427 |
+
start_offset = self.offset - self.flat_groups_offset[group_id]
|
| 428 |
+
end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
|
| 429 |
+
pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
|
| 430 |
+
|
| 431 |
+
# collect weights from all ranks
|
| 432 |
+
pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
|
| 433 |
+
param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
|
| 434 |
+
return param
|
| 435 |
+
|
| 436 |
+
|
| 437 |
+
def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
| 438 |
+
param_shapes = zero_model_states[0].param_shapes
|
| 439 |
+
avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
|
| 440 |
+
|
| 441 |
+
# Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
|
| 442 |
+
# param, re-consolidating each param, while dealing with padding if any
|
| 443 |
+
|
| 444 |
+
# merge list of dicts, preserving order
|
| 445 |
+
param_shapes = {k: v for d in param_shapes for k, v in d.items()}
|
| 446 |
+
|
| 447 |
+
if debug:
|
| 448 |
+
for i in range(world_size):
|
| 449 |
+
print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
|
| 450 |
+
|
| 451 |
+
wanted_params = len(param_shapes)
|
| 452 |
+
wanted_numel = sum(shape.numel() for shape in param_shapes.values())
|
| 453 |
+
# not asserting if there is a mismatch due to possible padding
|
| 454 |
+
avail_numel = fp32_flat_groups[0].numel() * world_size
|
| 455 |
+
print(f"Trainable params: Have {avail_numel} numels to process.")
|
| 456 |
+
print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
|
| 457 |
+
|
| 458 |
+
# params
|
| 459 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
| 460 |
+
# out-of-core computing solution
|
| 461 |
+
offset = 0
|
| 462 |
+
total_numel = 0
|
| 463 |
+
total_params = 0
|
| 464 |
+
flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
|
| 465 |
+
for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
|
| 466 |
+
unpartitioned_numel = shape.numel()
|
| 467 |
+
total_numel += unpartitioned_numel
|
| 468 |
+
total_params += 1
|
| 469 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
| 470 |
+
|
| 471 |
+
if debug:
|
| 472 |
+
print(
|
| 473 |
+
f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
| 474 |
+
)
|
| 475 |
+
|
| 476 |
+
# memory efficient tensor
|
| 477 |
+
tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
|
| 478 |
+
state_dict[name] = tensor
|
| 479 |
+
offset += partitioned_numel
|
| 480 |
+
|
| 481 |
+
offset *= world_size
|
| 482 |
+
|
| 483 |
+
# Sanity check
|
| 484 |
+
if offset != avail_numel:
|
| 485 |
+
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
| 486 |
+
|
| 487 |
+
print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
|
| 488 |
+
|
| 489 |
+
|
| 490 |
+
def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
| 491 |
+
exclude_frozen_parameters):
|
| 492 |
+
state_dict = OrderedDict()
|
| 493 |
+
|
| 494 |
+
# buffers
|
| 495 |
+
buffers = zero_model_states[0].buffers
|
| 496 |
+
state_dict.update(buffers)
|
| 497 |
+
if debug:
|
| 498 |
+
print(f"added {len(buffers)} buffers")
|
| 499 |
+
|
| 500 |
+
if not exclude_frozen_parameters:
|
| 501 |
+
_zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
|
| 502 |
+
|
| 503 |
+
_zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
| 504 |
+
|
| 505 |
+
# recover shared parameters
|
| 506 |
+
for pair in zero_model_states[0].shared_params:
|
| 507 |
+
if pair[1] in state_dict:
|
| 508 |
+
state_dict[pair[0]] = state_dict[pair[1]]
|
| 509 |
+
|
| 510 |
+
return state_dict
|
| 511 |
+
|
| 512 |
+
|
| 513 |
+
def to_torch_tensor(state_dict, return_empty_tensor=False):
|
| 514 |
+
"""
|
| 515 |
+
Convert state_dict of GatheredTensor to torch tensor
|
| 516 |
+
"""
|
| 517 |
+
torch_state_dict = {}
|
| 518 |
+
converted_tensors = {}
|
| 519 |
+
for name, tensor in state_dict.items():
|
| 520 |
+
tensor_id = id(tensor)
|
| 521 |
+
if tensor_id in converted_tensors: # shared tensors
|
| 522 |
+
shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
|
| 523 |
+
torch_state_dict[name] = shared_tensor
|
| 524 |
+
else:
|
| 525 |
+
converted_tensors[tensor_id] = name
|
| 526 |
+
if return_empty_tensor:
|
| 527 |
+
torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
|
| 528 |
+
else:
|
| 529 |
+
torch_state_dict[name] = tensor.contiguous()
|
| 530 |
+
return torch_state_dict
|
| 531 |
+
|
| 532 |
+
|
| 533 |
+
def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
|
| 534 |
+
tag=None,
|
| 535 |
+
exclude_frozen_parameters=False,
|
| 536 |
+
lazy_mode=False):
|
| 537 |
+
"""
|
| 538 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
|
| 539 |
+
``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
|
| 540 |
+
via a model hub.
|
| 541 |
+
|
| 542 |
+
Args:
|
| 543 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder
|
| 544 |
+
- ``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``
|
| 545 |
+
- ``exclude_frozen_parameters``: exclude frozen parameters
|
| 546 |
+
- ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
|
| 547 |
+
Convert the pesduo tensor to torch tensor by ``.contiguous()``
|
| 548 |
+
|
| 549 |
+
Returns:
|
| 550 |
+
- pytorch ``state_dict``
|
| 551 |
+
|
| 552 |
+
A typical usage might be ::
|
| 553 |
+
|
| 554 |
+
from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
|
| 555 |
+
# do the training and checkpoint saving
|
| 556 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
|
| 557 |
+
model = model.cpu() # move to cpu
|
| 558 |
+
model.load_state_dict(state_dict)
|
| 559 |
+
# submit to model hub or save the model to share with others
|
| 560 |
+
|
| 561 |
+
In this example the ``model`` will no longer be usable in the deepspeed context of the same
|
| 562 |
+
application. i.e. you will need to re-initialize the deepspeed engine, since
|
| 563 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
| 564 |
+
|
| 565 |
+
If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
|
| 566 |
+
|
| 567 |
+
Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
|
| 568 |
+
You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
|
| 569 |
+
the checkpoint. Or you can load state_dict in lazy mode ::
|
| 570 |
+
|
| 571 |
+
from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
|
| 572 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
|
| 573 |
+
for name, lazy_tensor in state_dict.item():
|
| 574 |
+
tensor = lazy_tensor.contiguous() # to cpu
|
| 575 |
+
print(name, tensor)
|
| 576 |
+
# del tensor to release memory if it no longer in use
|
| 577 |
+
"""
|
| 578 |
+
if tag is None:
|
| 579 |
+
latest_path = os.path.join(checkpoint_dir, 'latest')
|
| 580 |
+
if os.path.isfile(latest_path):
|
| 581 |
+
with open(latest_path, 'r') as fd:
|
| 582 |
+
tag = fd.read().strip()
|
| 583 |
+
else:
|
| 584 |
+
raise ValueError(f"Unable to find 'latest' file at {latest_path}")
|
| 585 |
+
|
| 586 |
+
ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
|
| 587 |
+
|
| 588 |
+
if not os.path.isdir(ds_checkpoint_dir):
|
| 589 |
+
raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
|
| 590 |
+
|
| 591 |
+
state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
|
| 592 |
+
if lazy_mode:
|
| 593 |
+
return state_dict
|
| 594 |
+
else:
|
| 595 |
+
return to_torch_tensor(state_dict)
|
| 596 |
+
|
| 597 |
+
|
| 598 |
+
def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
|
| 599 |
+
output_dir,
|
| 600 |
+
max_shard_size="5GB",
|
| 601 |
+
safe_serialization=False,
|
| 602 |
+
tag=None,
|
| 603 |
+
exclude_frozen_parameters=False):
|
| 604 |
+
"""
|
| 605 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
|
| 606 |
+
loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
|
| 607 |
+
|
| 608 |
+
Args:
|
| 609 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
| 610 |
+
- ``output_dir``: directory to the pytorch fp32 state_dict output files
|
| 611 |
+
- ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
|
| 612 |
+
- ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
|
| 613 |
+
- ``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``
|
| 614 |
+
- ``exclude_frozen_parameters``: exclude frozen parameters
|
| 615 |
+
"""
|
| 616 |
+
|
| 617 |
+
# Dependency pre-check
|
| 618 |
+
if safe_serialization:
|
| 619 |
+
try:
|
| 620 |
+
from safetensors.torch import save_file
|
| 621 |
+
except ImportError:
|
| 622 |
+
print('If you want to use `safe_serialization`, please `pip install safetensors`')
|
| 623 |
+
raise
|
| 624 |
+
if max_shard_size is not None:
|
| 625 |
+
try:
|
| 626 |
+
from huggingface_hub import split_torch_state_dict_into_shards
|
| 627 |
+
except ImportError:
|
| 628 |
+
print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
|
| 629 |
+
raise
|
| 630 |
+
|
| 631 |
+
# Convert zero checkpoint to state_dict
|
| 632 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
|
| 633 |
+
tag,
|
| 634 |
+
exclude_frozen_parameters,
|
| 635 |
+
lazy_mode=True)
|
| 636 |
+
|
| 637 |
+
# Shard the model if it is too big.
|
| 638 |
+
weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
|
| 639 |
+
if max_shard_size is not None:
|
| 640 |
+
filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
|
| 641 |
+
# an memory-efficient approach for sharding
|
| 642 |
+
empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
|
| 643 |
+
state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
|
| 644 |
+
filename_pattern=filename_pattern,
|
| 645 |
+
max_shard_size=max_shard_size)
|
| 646 |
+
else:
|
| 647 |
+
from collections import namedtuple
|
| 648 |
+
StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
|
| 649 |
+
state_dict_split = StateDictSplit(is_sharded=False,
|
| 650 |
+
filename_to_tensors={weights_name: list(state_dict.keys())})
|
| 651 |
+
|
| 652 |
+
# Save the model by shard
|
| 653 |
+
os.makedirs(output_dir, exist_ok=True)
|
| 654 |
+
filename_to_tensors = state_dict_split.filename_to_tensors.items()
|
| 655 |
+
for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
|
| 656 |
+
shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
|
| 657 |
+
shard_state_dict = to_torch_tensor(shard_state_dict)
|
| 658 |
+
output_path = os.path.join(output_dir, shard_file)
|
| 659 |
+
if safe_serialization:
|
| 660 |
+
save_file(shard_state_dict, output_path, metadata={"format": "pt"})
|
| 661 |
+
else:
|
| 662 |
+
torch.save(shard_state_dict, output_path)
|
| 663 |
+
# release the memory of current shard
|
| 664 |
+
for tensor_name in list(shard_state_dict.keys()):
|
| 665 |
+
del state_dict[tensor_name]
|
| 666 |
+
del shard_state_dict[tensor_name]
|
| 667 |
+
del shard_state_dict
|
| 668 |
+
gc.collect()
|
| 669 |
+
|
| 670 |
+
# Save index if sharded
|
| 671 |
+
if state_dict_split.is_sharded:
|
| 672 |
+
index = {
|
| 673 |
+
"metadata": state_dict_split.metadata,
|
| 674 |
+
"weight_map": state_dict_split.tensor_to_filename,
|
| 675 |
+
}
|
| 676 |
+
save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
|
| 677 |
+
save_index_file = os.path.join(output_dir, save_index_file)
|
| 678 |
+
with open(save_index_file, "w", encoding="utf-8") as f:
|
| 679 |
+
content = json.dumps(index, indent=2, sort_keys=True) + "\n"
|
| 680 |
+
f.write(content)
|
| 681 |
+
|
| 682 |
+
|
| 683 |
+
def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
|
| 684 |
+
"""
|
| 685 |
+
1. Put the provided model to cpu
|
| 686 |
+
2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
|
| 687 |
+
3. Load it into the provided model
|
| 688 |
+
|
| 689 |
+
Args:
|
| 690 |
+
- ``model``: the model object to update
|
| 691 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
| 692 |
+
- ``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``
|
| 693 |
+
|
| 694 |
+
Returns:
|
| 695 |
+
- ``model`: modified model
|
| 696 |
+
|
| 697 |
+
Make sure you have plenty of CPU memory available before you call this function. If you don't
|
| 698 |
+
have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
|
| 699 |
+
conveniently placed for you in the checkpoint folder.
|
| 700 |
+
|
| 701 |
+
A typical usage might be ::
|
| 702 |
+
|
| 703 |
+
from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
|
| 704 |
+
model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
|
| 705 |
+
# submit to model hub or save the model to share with others
|
| 706 |
+
|
| 707 |
+
Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
|
| 708 |
+
of the same application. i.e. you will need to re-initialize the deepspeed engine, since
|
| 709 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
| 710 |
+
|
| 711 |
+
"""
|
| 712 |
+
logger.info("Extracting fp32 weights")
|
| 713 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
| 714 |
+
|
| 715 |
+
logger.info("Overwriting model with fp32 weights")
|
| 716 |
+
model = model.cpu()
|
| 717 |
+
model.load_state_dict(state_dict, strict=False)
|
| 718 |
+
|
| 719 |
+
return model
|
| 720 |
+
|
| 721 |
+
|
| 722 |
+
if __name__ == "__main__":
|
| 723 |
+
parser = argparse.ArgumentParser()
|
| 724 |
+
parser.add_argument("checkpoint_dir",
|
| 725 |
+
type=str,
|
| 726 |
+
help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
|
| 727 |
+
parser.add_argument("output_dir",
|
| 728 |
+
type=str,
|
| 729 |
+
help="directory to the pytorch fp32 state_dict output files"
|
| 730 |
+
"(e.g. path/checkpoint-12-output/)")
|
| 731 |
+
parser.add_argument(
|
| 732 |
+
"--max_shard_size",
|
| 733 |
+
type=str,
|
| 734 |
+
default="5GB",
|
| 735 |
+
help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
|
| 736 |
+
"lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
|
| 737 |
+
"We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
|
| 738 |
+
"without CPU OOM issues.")
|
| 739 |
+
parser.add_argument(
|
| 740 |
+
"--safe_serialization",
|
| 741 |
+
default=False,
|
| 742 |
+
action='store_true',
|
| 743 |
+
help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
|
| 744 |
+
parser.add_argument("-t",
|
| 745 |
+
"--tag",
|
| 746 |
+
type=str,
|
| 747 |
+
default=None,
|
| 748 |
+
help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
|
| 749 |
+
parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
|
| 750 |
+
parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
|
| 751 |
+
args = parser.parse_args()
|
| 752 |
+
|
| 753 |
+
debug = args.debug
|
| 754 |
+
|
| 755 |
+
convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
|
| 756 |
+
args.output_dir,
|
| 757 |
+
max_shard_size=args.max_shard_size,
|
| 758 |
+
safe_serialization=args.safe_serialization,
|
| 759 |
+
tag=args.tag,
|
| 760 |
+
exclude_frozen_parameters=args.exclude_frozen_parameters)
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/args.json
ADDED
|
@@ -0,0 +1,376 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"output_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_v9_1_a800_10pct/v0-20260602-181529",
|
| 3 |
+
"per_device_train_batch_size": 1,
|
| 4 |
+
"num_train_epochs": 3.0,
|
| 5 |
+
"max_steps": 505,
|
| 6 |
+
"learning_rate": 1e-05,
|
| 7 |
+
"lr_scheduler_type": "cosine",
|
| 8 |
+
"lr_scheduler_kwargs": null,
|
| 9 |
+
"warmup_steps": 0,
|
| 10 |
+
"optim": "adamw_torch_fused",
|
| 11 |
+
"optim_args": null,
|
| 12 |
+
"weight_decay": 0.1,
|
| 13 |
+
"adam_beta1": 0.9,
|
| 14 |
+
"adam_beta2": 0.95,
|
| 15 |
+
"adam_epsilon": 1e-08,
|
| 16 |
+
"optim_target_modules": null,
|
| 17 |
+
"gradient_accumulation_steps": 8,
|
| 18 |
+
"average_tokens_across_devices": true,
|
| 19 |
+
"max_grad_norm": 1.0,
|
| 20 |
+
"label_smoothing_factor": 0.0,
|
| 21 |
+
"bf16": true,
|
| 22 |
+
"fp16": false,
|
| 23 |
+
"bf16_full_eval": false,
|
| 24 |
+
"fp16_full_eval": false,
|
| 25 |
+
"tf32": null,
|
| 26 |
+
"gradient_checkpointing": true,
|
| 27 |
+
"gradient_checkpointing_kwargs": "{\"use_reentrant\": false}",
|
| 28 |
+
"torch_compile": false,
|
| 29 |
+
"torch_compile_backend": null,
|
| 30 |
+
"torch_compile_mode": null,
|
| 31 |
+
"use_liger_kernel": false,
|
| 32 |
+
"liger_kernel_config": null,
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"neftune_noise_alpha": null,
|
| 35 |
+
"torch_empty_cache_steps": null,
|
| 36 |
+
"auto_find_batch_size": false,
|
| 37 |
+
"logging_strategy": "steps",
|
| 38 |
+
"logging_steps": 1,
|
| 39 |
+
"logging_first_step": true,
|
| 40 |
+
"log_on_each_node": true,
|
| 41 |
+
"logging_nan_inf_filter": true,
|
| 42 |
+
"include_num_input_tokens_seen": false,
|
| 43 |
+
"log_level": "passive",
|
| 44 |
+
"log_level_replica": "warning",
|
| 45 |
+
"disable_tqdm": null,
|
| 46 |
+
"report_to": [
|
| 47 |
+
"none"
|
| 48 |
+
],
|
| 49 |
+
"run_name": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_v9_1_a800_10pct/v0-20260602-181529",
|
| 50 |
+
"project": "huggingface",
|
| 51 |
+
"trackio_space_id": "trackio",
|
| 52 |
+
"eval_strategy": "no",
|
| 53 |
+
"eval_steps": 200.0,
|
| 54 |
+
"eval_delay": 0,
|
| 55 |
+
"per_device_eval_batch_size": 1,
|
| 56 |
+
"prediction_loss_only": false,
|
| 57 |
+
"eval_on_start": false,
|
| 58 |
+
"eval_do_concat_batches": true,
|
| 59 |
+
"eval_use_gather_object": false,
|
| 60 |
+
"eval_accumulation_steps": null,
|
| 61 |
+
"include_for_metrics": [],
|
| 62 |
+
"batch_eval_metrics": false,
|
| 63 |
+
"save_only_model": false,
|
| 64 |
+
"save_strategy": "steps",
|
| 65 |
+
"save_steps": 200.0,
|
| 66 |
+
"save_on_each_node": false,
|
| 67 |
+
"save_total_limit": 1,
|
| 68 |
+
"enable_jit_checkpoint": false,
|
| 69 |
+
"push_to_hub": false,
|
| 70 |
+
"hub_token": null,
|
| 71 |
+
"hub_private_repo": null,
|
| 72 |
+
"hub_model_id": null,
|
| 73 |
+
"hub_strategy": "every_save",
|
| 74 |
+
"hub_always_push": false,
|
| 75 |
+
"hub_revision": null,
|
| 76 |
+
"load_best_model_at_end": false,
|
| 77 |
+
"metric_for_best_model": "loss",
|
| 78 |
+
"greater_is_better": false,
|
| 79 |
+
"ignore_data_skip": false,
|
| 80 |
+
"restore_callback_states_from_checkpoint": false,
|
| 81 |
+
"full_determinism": false,
|
| 82 |
+
"seed": 42,
|
| 83 |
+
"data_seed": 42,
|
| 84 |
+
"use_cpu": false,
|
| 85 |
+
"accelerator_config": "{\"dispatch_batches\": false}",
|
| 86 |
+
"parallelism_config": null,
|
| 87 |
+
"dataloader_drop_last": false,
|
| 88 |
+
"dataloader_num_workers": 6,
|
| 89 |
+
"dataloader_pin_memory": true,
|
| 90 |
+
"dataloader_persistent_workers": true,
|
| 91 |
+
"dataloader_prefetch_factor": 4,
|
| 92 |
+
"remove_unused_columns": true,
|
| 93 |
+
"label_names": null,
|
| 94 |
+
"train_sampling_strategy": "random",
|
| 95 |
+
"length_column_name": "length",
|
| 96 |
+
"ddp_find_unused_parameters": null,
|
| 97 |
+
"ddp_bucket_cap_mb": null,
|
| 98 |
+
"ddp_broadcast_buffers": null,
|
| 99 |
+
"ddp_backend": null,
|
| 100 |
+
"ddp_timeout": 7200,
|
| 101 |
+
"fsdp": [],
|
| 102 |
+
"fsdp_config": null,
|
| 103 |
+
"deepspeed": {
|
| 104 |
+
"fp16": {
|
| 105 |
+
"enabled": "auto",
|
| 106 |
+
"loss_scale": 0,
|
| 107 |
+
"loss_scale_window": 1000,
|
| 108 |
+
"initial_scale_power": 16,
|
| 109 |
+
"hysteresis": 2,
|
| 110 |
+
"min_loss_scale": 1
|
| 111 |
+
},
|
| 112 |
+
"bf16": {
|
| 113 |
+
"enabled": "auto"
|
| 114 |
+
},
|
| 115 |
+
"zero_optimization": {
|
| 116 |
+
"stage": 2,
|
| 117 |
+
"offload_optimizer": {
|
| 118 |
+
"device": "none",
|
| 119 |
+
"pin_memory": true
|
| 120 |
+
},
|
| 121 |
+
"allgather_partitions": true,
|
| 122 |
+
"allgather_bucket_size": 200000000.0,
|
| 123 |
+
"overlap_comm": false,
|
| 124 |
+
"reduce_scatter": true,
|
| 125 |
+
"reduce_bucket_size": 200000000.0,
|
| 126 |
+
"contiguous_gradients": true
|
| 127 |
+
},
|
| 128 |
+
"gradient_accumulation_steps": "auto",
|
| 129 |
+
"gradient_clipping": "auto",
|
| 130 |
+
"steps_per_print": 2000,
|
| 131 |
+
"train_batch_size": "auto",
|
| 132 |
+
"train_micro_batch_size_per_gpu": "auto",
|
| 133 |
+
"wall_clock_breakdown": false
|
| 134 |
+
},
|
| 135 |
+
"debug": null,
|
| 136 |
+
"skip_memory_metrics": true,
|
| 137 |
+
"do_train": false,
|
| 138 |
+
"do_eval": false,
|
| 139 |
+
"do_predict": false,
|
| 140 |
+
"resume_from_checkpoint": null,
|
| 141 |
+
"warmup_ratio": 0.05,
|
| 142 |
+
"logging_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s2_v9_1_a800_10pct/v0-20260602-181529/runs",
|
| 143 |
+
"local_rank": 0,
|
| 144 |
+
"sortish_sampler": false,
|
| 145 |
+
"predict_with_generate": false,
|
| 146 |
+
"generation_max_length": null,
|
| 147 |
+
"generation_num_beams": null,
|
| 148 |
+
"generation_config": null,
|
| 149 |
+
"tuner_backend": "peft",
|
| 150 |
+
"vit_gradient_checkpointing": true,
|
| 151 |
+
"router_aux_loss_coef": 0.0,
|
| 152 |
+
"enable_dft_loss": false,
|
| 153 |
+
"enable_channel_loss": false,
|
| 154 |
+
"safe_serialization": true,
|
| 155 |
+
"max_shard_size": "5GB",
|
| 156 |
+
"check_model": true,
|
| 157 |
+
"acc_strategy": "token",
|
| 158 |
+
"train_dataloader_shuffle": true,
|
| 159 |
+
"group_by_length": false,
|
| 160 |
+
"max_epochs": null,
|
| 161 |
+
"aligner_lr": null,
|
| 162 |
+
"vit_lr": 1e-06,
|
| 163 |
+
"use_logits_to_keep": null,
|
| 164 |
+
"ds3_gather_for_generation": true,
|
| 165 |
+
"resume_only_model": false,
|
| 166 |
+
"optimizer": null,
|
| 167 |
+
"loss_type": null,
|
| 168 |
+
"eval_metric": null,
|
| 169 |
+
"callbacks": [],
|
| 170 |
+
"early_stop_interval": null,
|
| 171 |
+
"eval_use_evalscope": false,
|
| 172 |
+
"eval_dataset": [],
|
| 173 |
+
"eval_dataset_args": null,
|
| 174 |
+
"eval_limit": null,
|
| 175 |
+
"eval_generation_config": null,
|
| 176 |
+
"extra_eval_args": null,
|
| 177 |
+
"tuner_type": "full",
|
| 178 |
+
"use_galore": false,
|
| 179 |
+
"galore_target_modules": null,
|
| 180 |
+
"galore_rank": 128,
|
| 181 |
+
"galore_update_proj_gap": 50,
|
| 182 |
+
"galore_scale": 1.0,
|
| 183 |
+
"galore_proj_type": "std",
|
| 184 |
+
"galore_optim_per_parameter": false,
|
| 185 |
+
"galore_with_embedding": false,
|
| 186 |
+
"galore_quantization": false,
|
| 187 |
+
"galore_proj_quant": false,
|
| 188 |
+
"galore_proj_bits": 4,
|
| 189 |
+
"galore_proj_group_size": 256,
|
| 190 |
+
"galore_cos_threshold": 0.4,
|
| 191 |
+
"galore_gamma_proj": 2,
|
| 192 |
+
"galore_queue_size": 5,
|
| 193 |
+
"lisa_activated_layers": 0,
|
| 194 |
+
"lisa_step_interval": 20,
|
| 195 |
+
"use_flash_ckpt": false,
|
| 196 |
+
"use_ray": false,
|
| 197 |
+
"ray_exp_name": null,
|
| 198 |
+
"device_groups": null,
|
| 199 |
+
"model": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_v9_1_a800/v0-20260602-115055/checkpoint-2181",
|
| 200 |
+
"model_type": "llava_qwen3_5vit_qwen3",
|
| 201 |
+
"model_revision": null,
|
| 202 |
+
"task_type": "causal_lm",
|
| 203 |
+
"torch_dtype": "bfloat16",
|
| 204 |
+
"attn_impl": "flash_attn",
|
| 205 |
+
"experts_impl": null,
|
| 206 |
+
"new_special_tokens": [],
|
| 207 |
+
"num_labels": null,
|
| 208 |
+
"problem_type": null,
|
| 209 |
+
"rope_scaling": null,
|
| 210 |
+
"device_map": null,
|
| 211 |
+
"max_memory": {},
|
| 212 |
+
"max_model_len": null,
|
| 213 |
+
"local_repo_path": null,
|
| 214 |
+
"init_strategy": null,
|
| 215 |
+
"template": "llava_qwen3_5vit_qwen3",
|
| 216 |
+
"system": null,
|
| 217 |
+
"max_length": 16384,
|
| 218 |
+
"truncation_strategy": "delete",
|
| 219 |
+
"max_pixels": null,
|
| 220 |
+
"agent_template": null,
|
| 221 |
+
"norm_bbox": null,
|
| 222 |
+
"use_chat_template": true,
|
| 223 |
+
"padding_side": "right",
|
| 224 |
+
"padding_free": true,
|
| 225 |
+
"loss_scale": "default",
|
| 226 |
+
"sequence_parallel_size": 1,
|
| 227 |
+
"template_backend": "swift",
|
| 228 |
+
"response_prefix": null,
|
| 229 |
+
"enable_thinking": null,
|
| 230 |
+
"add_non_thinking_prefix": true,
|
| 231 |
+
"dataset": [],
|
| 232 |
+
"val_dataset": [],
|
| 233 |
+
"cached_dataset": [
|
| 234 |
+
"/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/image_10pct/train",
|
| 235 |
+
"/share/m2v_intern_v3/wangjunjie09/VisionEncoder/data/vmllm_cached/qwen3vit/video_10pct/train"
|
| 236 |
+
],
|
| 237 |
+
"cached_val_dataset": [],
|
| 238 |
+
"split_dataset_ratio": 0.0,
|
| 239 |
+
"dataset_num_proc": 16,
|
| 240 |
+
"load_from_cache_file": false,
|
| 241 |
+
"dataset_shuffle": true,
|
| 242 |
+
"val_dataset_shuffle": false,
|
| 243 |
+
"streaming": false,
|
| 244 |
+
"interleave_prob": null,
|
| 245 |
+
"stopping_strategy": "first_exhausted",
|
| 246 |
+
"shuffle_buffer_size": 1000,
|
| 247 |
+
"download_mode": "reuse_dataset_if_exists",
|
| 248 |
+
"columns": {},
|
| 249 |
+
"strict": false,
|
| 250 |
+
"model_name": null,
|
| 251 |
+
"model_author": null,
|
| 252 |
+
"custom_dataset_info": [],
|
| 253 |
+
"quant_method": null,
|
| 254 |
+
"quant_bits": null,
|
| 255 |
+
"hqq_axis": null,
|
| 256 |
+
"bnb_4bit_compute_dtype": "bfloat16",
|
| 257 |
+
"bnb_4bit_quant_type": "nf4",
|
| 258 |
+
"bnb_4bit_use_double_quant": true,
|
| 259 |
+
"bnb_4bit_quant_storage": null,
|
| 260 |
+
"max_new_tokens": 64,
|
| 261 |
+
"temperature": 0.0,
|
| 262 |
+
"top_k": null,
|
| 263 |
+
"top_p": null,
|
| 264 |
+
"repetition_penalty": null,
|
| 265 |
+
"num_beams": 1,
|
| 266 |
+
"stream": false,
|
| 267 |
+
"stop_words": [],
|
| 268 |
+
"logprobs": false,
|
| 269 |
+
"top_logprobs": null,
|
| 270 |
+
"structured_outputs_regex": null,
|
| 271 |
+
"adapters": [],
|
| 272 |
+
"external_plugins": [
|
| 273 |
+
"ms-swift/video_mllm/model_plugin_qwen3_5vit.py",
|
| 274 |
+
"ms-swift/video_mllm/dataset_plugin.py"
|
| 275 |
+
],
|
| 276 |
+
"custom_register_path": [],
|
| 277 |
+
"model_kwargs": {},
|
| 278 |
+
"load_args": false,
|
| 279 |
+
"load_data_args": false,
|
| 280 |
+
"packing": true,
|
| 281 |
+
"packing_length": 16384,
|
| 282 |
+
"packing_num_proc": 1,
|
| 283 |
+
"lazy_tokenize": false,
|
| 284 |
+
"use_hf": true,
|
| 285 |
+
"ignore_args_error": false,
|
| 286 |
+
"use_swift_lora": false,
|
| 287 |
+
"freeze_parameters": [],
|
| 288 |
+
"freeze_parameters_regex": null,
|
| 289 |
+
"freeze_parameters_ratio": 0.0,
|
| 290 |
+
"trainable_parameters": [
|
| 291 |
+
"model.multi_modal_projector"
|
| 292 |
+
],
|
| 293 |
+
"trainable_parameters_regex": null,
|
| 294 |
+
"freeze_llm": false,
|
| 295 |
+
"freeze_vit": false,
|
| 296 |
+
"freeze_aligner": false,
|
| 297 |
+
"target_modules": [
|
| 298 |
+
"all-linear"
|
| 299 |
+
],
|
| 300 |
+
"target_regex": null,
|
| 301 |
+
"target_parameters": null,
|
| 302 |
+
"modules_to_save": [],
|
| 303 |
+
"lora_rank": 8,
|
| 304 |
+
"lora_alpha": 32,
|
| 305 |
+
"lora_dropout": 0.05,
|
| 306 |
+
"lora_bias": "none",
|
| 307 |
+
"lora_dtype": null,
|
| 308 |
+
"lorap_lr_ratio": null,
|
| 309 |
+
"use_rslora": false,
|
| 310 |
+
"use_dora": false,
|
| 311 |
+
"lora_ga_batch_size": 2,
|
| 312 |
+
"lora_ga_iters": 2,
|
| 313 |
+
"lora_ga_max_length": 1024,
|
| 314 |
+
"lora_ga_direction": "ArB2r",
|
| 315 |
+
"lora_ga_scale": "stable",
|
| 316 |
+
"lora_ga_stable_gamma": 16,
|
| 317 |
+
"init_weights": true,
|
| 318 |
+
"fourier_n_frequency": 2000,
|
| 319 |
+
"fourier_scaling": 300.0,
|
| 320 |
+
"boft_block_size": 4,
|
| 321 |
+
"boft_block_num": 0,
|
| 322 |
+
"boft_n_butterfly_factor": 1,
|
| 323 |
+
"boft_dropout": 0.0,
|
| 324 |
+
"vera_rank": 256,
|
| 325 |
+
"vera_projection_prng_key": 0,
|
| 326 |
+
"vera_dropout": 0.0,
|
| 327 |
+
"vera_d_initial": 0.1,
|
| 328 |
+
"adapter_act": "gelu",
|
| 329 |
+
"adapter_length": 128,
|
| 330 |
+
"adalora_target_r": 8,
|
| 331 |
+
"adalora_init_r": 12,
|
| 332 |
+
"adalora_tinit": 0,
|
| 333 |
+
"adalora_tfinal": 0,
|
| 334 |
+
"adalora_deltaT": 1,
|
| 335 |
+
"adalora_beta1": 0.85,
|
| 336 |
+
"adalora_beta2": 0.85,
|
| 337 |
+
"adalora_orth_reg_weight": 0.5,
|
| 338 |
+
"llamapro_num_new_blocks": 4,
|
| 339 |
+
"llamapro_num_groups": null,
|
| 340 |
+
"reft_layer_key": null,
|
| 341 |
+
"reft_layers": null,
|
| 342 |
+
"reft_rank": 4,
|
| 343 |
+
"reft_intervention_type": "LoreftIntervention",
|
| 344 |
+
"reft_args": null,
|
| 345 |
+
"swanlab_token": null,
|
| 346 |
+
"swanlab_project": "ms-swift",
|
| 347 |
+
"swanlab_workspace": null,
|
| 348 |
+
"swanlab_exp_name": null,
|
| 349 |
+
"swanlab_notification_method": null,
|
| 350 |
+
"swanlab_webhook_url": null,
|
| 351 |
+
"swanlab_secret": null,
|
| 352 |
+
"swanlab_sender_email": null,
|
| 353 |
+
"swanlab_receiver_email": null,
|
| 354 |
+
"swanlab_smtp_server": null,
|
| 355 |
+
"swanlab_smtp_port": null,
|
| 356 |
+
"swanlab_email_language": "zh",
|
| 357 |
+
"swanlab_mode": "cloud",
|
| 358 |
+
"add_version": true,
|
| 359 |
+
"create_checkpoint_symlink": false,
|
| 360 |
+
"zero_hpz_partition_size": null,
|
| 361 |
+
"deepspeed_autotp_size": null,
|
| 362 |
+
"swift_version": "4.1.1",
|
| 363 |
+
"ckpt_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_v9_1_a800/v0-20260602-115055/checkpoint-2181",
|
| 364 |
+
"rank": 0,
|
| 365 |
+
"global_world_size": 8,
|
| 366 |
+
"local_world_size": 8,
|
| 367 |
+
"model_suffix": "checkpoint-2181",
|
| 368 |
+
"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)",
|
| 369 |
+
"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'])",
|
| 370 |
+
"model_dir": "/share/m2v_intern_v3/wangjunjie09/VisionEncoder/exps/video_mllm_swift/4b/s1_v9_1_a800/v0-20260602-115055/checkpoint-2181",
|
| 371 |
+
"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='')",
|
| 372 |
+
"_val_dataset_exists": false,
|
| 373 |
+
"hub": "<class 'swift.hub.hub.HFHub'>",
|
| 374 |
+
"evaluation_strategy": "steps",
|
| 375 |
+
"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)"
|
| 376 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/chat_template.jinja
ADDED
|
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{%- if tools %}
|
| 2 |
+
{{- '<|im_start|>system\n' }}
|
| 3 |
+
{%- if messages[0].role == 'system' %}
|
| 4 |
+
{{- messages[0].content + '\n\n' }}
|
| 5 |
+
{%- endif %}
|
| 6 |
+
{{- "# 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>" }}
|
| 7 |
+
{%- for tool in tools %}
|
| 8 |
+
{{- "\n" }}
|
| 9 |
+
{{- tool | tojson }}
|
| 10 |
+
{%- endfor %}
|
| 11 |
+
{{- "\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" }}
|
| 12 |
+
{%- else %}
|
| 13 |
+
{%- if messages[0].role == 'system' %}
|
| 14 |
+
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
| 15 |
+
{%- endif %}
|
| 16 |
+
{%- endif %}
|
| 17 |
+
{%- for message in messages %}
|
| 18 |
+
{%- if message.content is string %}
|
| 19 |
+
{%- set content = message.content %}
|
| 20 |
+
{%- else %}
|
| 21 |
+
{%- set content = '' %}
|
| 22 |
+
{%- endif %}
|
| 23 |
+
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
| 24 |
+
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
| 25 |
+
{%- elif message.role == "assistant" %}
|
| 26 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 27 |
+
{%- if message.tool_calls %}
|
| 28 |
+
{%- for tool_call in message.tool_calls %}
|
| 29 |
+
{%- if (loop.first and content) or (not loop.first) %}
|
| 30 |
+
{{- '\n' }}
|
| 31 |
+
{%- endif %}
|
| 32 |
+
{%- if tool_call.function %}
|
| 33 |
+
{%- set tool_call = tool_call.function %}
|
| 34 |
+
{%- endif %}
|
| 35 |
+
{{- '<tool_call>\n{"name": "' }}
|
| 36 |
+
{{- tool_call.name }}
|
| 37 |
+
{{- '", "arguments": ' }}
|
| 38 |
+
{%- if tool_call.arguments is string %}
|
| 39 |
+
{{- tool_call.arguments }}
|
| 40 |
+
{%- else %}
|
| 41 |
+
{{- tool_call.arguments | tojson }}
|
| 42 |
+
{%- endif %}
|
| 43 |
+
{{- '}\n</tool_call>' }}
|
| 44 |
+
{%- endfor %}
|
| 45 |
+
{%- endif %}
|
| 46 |
+
{{- '<|im_end|>\n' }}
|
| 47 |
+
{%- elif message.role == "tool" %}
|
| 48 |
+
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
| 49 |
+
{{- '<|im_start|>user' }}
|
| 50 |
+
{%- endif %}
|
| 51 |
+
{{- '\n<tool_response>\n' }}
|
| 52 |
+
{{- content }}
|
| 53 |
+
{{- '\n</tool_response>' }}
|
| 54 |
+
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
| 55 |
+
{{- '<|im_end|>\n' }}
|
| 56 |
+
{%- endif %}
|
| 57 |
+
{%- endif %}
|
| 58 |
+
{%- endfor %}
|
| 59 |
+
{%- if add_generation_prompt %}
|
| 60 |
+
{{- '<|im_start|>assistant\n' }}
|
| 61 |
+
{%- endif %}
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/chat_template.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"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 %}"
|
| 3 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/config.json
ADDED
|
@@ -0,0 +1,262 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"LlavaQwen3_5ViTForConditionalGeneration"
|
| 4 |
+
],
|
| 5 |
+
"bos_token_id": null,
|
| 6 |
+
"dtype": "bfloat16",
|
| 7 |
+
"eos_token_id": 151645,
|
| 8 |
+
"hidden_size": 2560,
|
| 9 |
+
"image_grid_pinpoints": [
|
| 10 |
+
[
|
| 11 |
+
384,
|
| 12 |
+
384
|
| 13 |
+
],
|
| 14 |
+
[
|
| 15 |
+
384,
|
| 16 |
+
768
|
| 17 |
+
],
|
| 18 |
+
[
|
| 19 |
+
384,
|
| 20 |
+
1152
|
| 21 |
+
],
|
| 22 |
+
[
|
| 23 |
+
384,
|
| 24 |
+
1536
|
| 25 |
+
],
|
| 26 |
+
[
|
| 27 |
+
384,
|
| 28 |
+
1920
|
| 29 |
+
],
|
| 30 |
+
[
|
| 31 |
+
384,
|
| 32 |
+
2304
|
| 33 |
+
],
|
| 34 |
+
[
|
| 35 |
+
768,
|
| 36 |
+
384
|
| 37 |
+
],
|
| 38 |
+
[
|
| 39 |
+
768,
|
| 40 |
+
768
|
| 41 |
+
],
|
| 42 |
+
[
|
| 43 |
+
768,
|
| 44 |
+
1152
|
| 45 |
+
],
|
| 46 |
+
[
|
| 47 |
+
768,
|
| 48 |
+
1536
|
| 49 |
+
],
|
| 50 |
+
[
|
| 51 |
+
768,
|
| 52 |
+
1920
|
| 53 |
+
],
|
| 54 |
+
[
|
| 55 |
+
768,
|
| 56 |
+
2304
|
| 57 |
+
],
|
| 58 |
+
[
|
| 59 |
+
1152,
|
| 60 |
+
384
|
| 61 |
+
],
|
| 62 |
+
[
|
| 63 |
+
1152,
|
| 64 |
+
768
|
| 65 |
+
],
|
| 66 |
+
[
|
| 67 |
+
1152,
|
| 68 |
+
1152
|
| 69 |
+
],
|
| 70 |
+
[
|
| 71 |
+
1152,
|
| 72 |
+
1536
|
| 73 |
+
],
|
| 74 |
+
[
|
| 75 |
+
1152,
|
| 76 |
+
1920
|
| 77 |
+
],
|
| 78 |
+
[
|
| 79 |
+
1152,
|
| 80 |
+
2304
|
| 81 |
+
],
|
| 82 |
+
[
|
| 83 |
+
1536,
|
| 84 |
+
384
|
| 85 |
+
],
|
| 86 |
+
[
|
| 87 |
+
1536,
|
| 88 |
+
768
|
| 89 |
+
],
|
| 90 |
+
[
|
| 91 |
+
1536,
|
| 92 |
+
1152
|
| 93 |
+
],
|
| 94 |
+
[
|
| 95 |
+
1536,
|
| 96 |
+
1536
|
| 97 |
+
],
|
| 98 |
+
[
|
| 99 |
+
1536,
|
| 100 |
+
1920
|
| 101 |
+
],
|
| 102 |
+
[
|
| 103 |
+
1536,
|
| 104 |
+
2304
|
| 105 |
+
],
|
| 106 |
+
[
|
| 107 |
+
1920,
|
| 108 |
+
384
|
| 109 |
+
],
|
| 110 |
+
[
|
| 111 |
+
1920,
|
| 112 |
+
768
|
| 113 |
+
],
|
| 114 |
+
[
|
| 115 |
+
1920,
|
| 116 |
+
1152
|
| 117 |
+
],
|
| 118 |
+
[
|
| 119 |
+
1920,
|
| 120 |
+
1536
|
| 121 |
+
],
|
| 122 |
+
[
|
| 123 |
+
1920,
|
| 124 |
+
1920
|
| 125 |
+
],
|
| 126 |
+
[
|
| 127 |
+
1920,
|
| 128 |
+
2304
|
| 129 |
+
],
|
| 130 |
+
[
|
| 131 |
+
2304,
|
| 132 |
+
384
|
| 133 |
+
],
|
| 134 |
+
[
|
| 135 |
+
2304,
|
| 136 |
+
768
|
| 137 |
+
],
|
| 138 |
+
[
|
| 139 |
+
2304,
|
| 140 |
+
1152
|
| 141 |
+
],
|
| 142 |
+
[
|
| 143 |
+
2304,
|
| 144 |
+
1536
|
| 145 |
+
],
|
| 146 |
+
[
|
| 147 |
+
2304,
|
| 148 |
+
1920
|
| 149 |
+
],
|
| 150 |
+
[
|
| 151 |
+
2304,
|
| 152 |
+
2304
|
| 153 |
+
]
|
| 154 |
+
],
|
| 155 |
+
"image_token_index": 151669,
|
| 156 |
+
"model_type": "llava_qwen3_5vit_qwen3",
|
| 157 |
+
"multimodal_projector_bias": true,
|
| 158 |
+
"pad_token_id": 151643,
|
| 159 |
+
"projector_hidden_act": "gelu",
|
| 160 |
+
"text_config": {
|
| 161 |
+
"_name_or_path": "/share/m2v_intern_v3/wangjunjie09/model_cache/huggingface/Qwen/Qwen3-4B-Instruct-2507",
|
| 162 |
+
"architectures": [
|
| 163 |
+
"Qwen3ForCausalLM"
|
| 164 |
+
],
|
| 165 |
+
"attention_bias": false,
|
| 166 |
+
"attention_dropout": 0.0,
|
| 167 |
+
"bos_token_id": 151643,
|
| 168 |
+
"dtype": "bfloat16",
|
| 169 |
+
"eos_token_id": 151645,
|
| 170 |
+
"head_dim": 128,
|
| 171 |
+
"hidden_act": "silu",
|
| 172 |
+
"hidden_size": 2560,
|
| 173 |
+
"initializer_range": 0.02,
|
| 174 |
+
"intermediate_size": 9728,
|
| 175 |
+
"layer_types": [
|
| 176 |
+
"full_attention",
|
| 177 |
+
"full_attention",
|
| 178 |
+
"full_attention",
|
| 179 |
+
"full_attention",
|
| 180 |
+
"full_attention",
|
| 181 |
+
"full_attention",
|
| 182 |
+
"full_attention",
|
| 183 |
+
"full_attention",
|
| 184 |
+
"full_attention",
|
| 185 |
+
"full_attention",
|
| 186 |
+
"full_attention",
|
| 187 |
+
"full_attention",
|
| 188 |
+
"full_attention",
|
| 189 |
+
"full_attention",
|
| 190 |
+
"full_attention",
|
| 191 |
+
"full_attention",
|
| 192 |
+
"full_attention",
|
| 193 |
+
"full_attention",
|
| 194 |
+
"full_attention",
|
| 195 |
+
"full_attention",
|
| 196 |
+
"full_attention",
|
| 197 |
+
"full_attention",
|
| 198 |
+
"full_attention",
|
| 199 |
+
"full_attention",
|
| 200 |
+
"full_attention",
|
| 201 |
+
"full_attention",
|
| 202 |
+
"full_attention",
|
| 203 |
+
"full_attention",
|
| 204 |
+
"full_attention",
|
| 205 |
+
"full_attention",
|
| 206 |
+
"full_attention",
|
| 207 |
+
"full_attention",
|
| 208 |
+
"full_attention",
|
| 209 |
+
"full_attention",
|
| 210 |
+
"full_attention",
|
| 211 |
+
"full_attention"
|
| 212 |
+
],
|
| 213 |
+
"max_position_embeddings": 262144,
|
| 214 |
+
"max_window_layers": 36,
|
| 215 |
+
"model_type": "qwen3",
|
| 216 |
+
"num_attention_heads": 32,
|
| 217 |
+
"num_hidden_layers": 36,
|
| 218 |
+
"num_key_value_heads": 8,
|
| 219 |
+
"pad_token_id": 151643,
|
| 220 |
+
"rms_norm_eps": 1e-06,
|
| 221 |
+
"rope_parameters": {
|
| 222 |
+
"rope_theta": 5000000,
|
| 223 |
+
"rope_type": "default"
|
| 224 |
+
},
|
| 225 |
+
"sliding_window": null,
|
| 226 |
+
"tie_word_embeddings": true,
|
| 227 |
+
"use_cache": false,
|
| 228 |
+
"use_sliding_window": false,
|
| 229 |
+
"vocab_size": 151936
|
| 230 |
+
},
|
| 231 |
+
"tie_word_embeddings": true,
|
| 232 |
+
"transformers_version": "5.5.4",
|
| 233 |
+
"use_cache": false,
|
| 234 |
+
"video_token_index": 151670,
|
| 235 |
+
"vision_aspect_ratio": "anyres_max_9",
|
| 236 |
+
"vision_config": {
|
| 237 |
+
"deepstack_visual_indexes": [],
|
| 238 |
+
"depth": 24,
|
| 239 |
+
"dtype": "bfloat16",
|
| 240 |
+
"hidden_act": "gelu_pytorch_tanh",
|
| 241 |
+
"hidden_size": 1024,
|
| 242 |
+
"image_size": 384,
|
| 243 |
+
"in_channels": 3,
|
| 244 |
+
"initializer_range": 0.02,
|
| 245 |
+
"intermediate_size": 4096,
|
| 246 |
+
"model_type": "qwen3_5",
|
| 247 |
+
"num_heads": 16,
|
| 248 |
+
"num_position_embeddings": 2304,
|
| 249 |
+
"out_hidden_size": 2048,
|
| 250 |
+
"patch_size": 16,
|
| 251 |
+
"spatial_merge_size": 2,
|
| 252 |
+
"temporal_patch_size": 2
|
| 253 |
+
},
|
| 254 |
+
"vision_feature_layer": -1,
|
| 255 |
+
"vision_feature_select_strategy": "full",
|
| 256 |
+
"auto_map": {
|
| 257 |
+
"AutoConfig": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTConfig",
|
| 258 |
+
"AutoModel": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
|
| 259 |
+
"AutoModelForCausalLM": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration",
|
| 260 |
+
"AutoModelForImageTextToText": "modeling_qwen3_5vit_qwen3.LlavaQwen3_5ViTForConditionalGeneration"
|
| 261 |
+
}
|
| 262 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/generation_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"bos_token_id": 151643,
|
| 4 |
+
"eos_token_id": [
|
| 5 |
+
151645,
|
| 6 |
+
151643
|
| 7 |
+
],
|
| 8 |
+
"output_attentions": false,
|
| 9 |
+
"output_hidden_states": false,
|
| 10 |
+
"transformers_version": "5.5.4",
|
| 11 |
+
"use_cache": true
|
| 12 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/latest
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
global_step505
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/model-00001-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:daecd183bea10424f8e65aad0a672571b19b72fd25499991ff5e494c2072a448
|
| 3 |
+
size 4987507408
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/model-00002-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c36c9a449fbcb3cecc3ae61966ed4c49f1346a1543193ba3b00bf848b20153bd
|
| 3 |
+
size 4466285248
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/model.safetensors.index.json
ADDED
|
@@ -0,0 +1,705 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metadata": {
|
| 3 |
+
"total_parameters": 4337895424,
|
| 4 |
+
"total_size": 9453703168
|
| 5 |
+
},
|
| 6 |
+
"weight_map": {
|
| 7 |
+
"lm_head.weight": "model-00001-of-00002.safetensors",
|
| 8 |
+
"model.image_newline": "model-00001-of-00002.safetensors",
|
| 9 |
+
"model.language_model.embed_tokens.weight": "model-00001-of-00002.safetensors",
|
| 10 |
+
"model.language_model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 11 |
+
"model.language_model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 12 |
+
"model.language_model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 13 |
+
"model.language_model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 14 |
+
"model.language_model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 15 |
+
"model.language_model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 16 |
+
"model.language_model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 17 |
+
"model.language_model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 18 |
+
"model.language_model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 19 |
+
"model.language_model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 20 |
+
"model.language_model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 21 |
+
"model.language_model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 22 |
+
"model.language_model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 23 |
+
"model.language_model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 24 |
+
"model.language_model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 25 |
+
"model.language_model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 26 |
+
"model.language_model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 27 |
+
"model.language_model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 28 |
+
"model.language_model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 29 |
+
"model.language_model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 30 |
+
"model.language_model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 31 |
+
"model.language_model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 32 |
+
"model.language_model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 33 |
+
"model.language_model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 34 |
+
"model.language_model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 35 |
+
"model.language_model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 36 |
+
"model.language_model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 37 |
+
"model.language_model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 38 |
+
"model.language_model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 39 |
+
"model.language_model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 40 |
+
"model.language_model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 41 |
+
"model.language_model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 42 |
+
"model.language_model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 43 |
+
"model.language_model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 44 |
+
"model.language_model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 45 |
+
"model.language_model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 46 |
+
"model.language_model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 47 |
+
"model.language_model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 48 |
+
"model.language_model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 49 |
+
"model.language_model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 50 |
+
"model.language_model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 51 |
+
"model.language_model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 52 |
+
"model.language_model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 53 |
+
"model.language_model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 54 |
+
"model.language_model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 55 |
+
"model.language_model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 56 |
+
"model.language_model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 57 |
+
"model.language_model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 58 |
+
"model.language_model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 59 |
+
"model.language_model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 60 |
+
"model.language_model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 61 |
+
"model.language_model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 62 |
+
"model.language_model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 63 |
+
"model.language_model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 64 |
+
"model.language_model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 65 |
+
"model.language_model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 66 |
+
"model.language_model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 67 |
+
"model.language_model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 68 |
+
"model.language_model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 69 |
+
"model.language_model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 70 |
+
"model.language_model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 71 |
+
"model.language_model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 72 |
+
"model.language_model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 73 |
+
"model.language_model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 74 |
+
"model.language_model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 75 |
+
"model.language_model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 76 |
+
"model.language_model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 77 |
+
"model.language_model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 78 |
+
"model.language_model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 79 |
+
"model.language_model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 80 |
+
"model.language_model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 81 |
+
"model.language_model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 82 |
+
"model.language_model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 83 |
+
"model.language_model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 84 |
+
"model.language_model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 85 |
+
"model.language_model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 86 |
+
"model.language_model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 87 |
+
"model.language_model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 88 |
+
"model.language_model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 89 |
+
"model.language_model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 90 |
+
"model.language_model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 91 |
+
"model.language_model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 92 |
+
"model.language_model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 93 |
+
"model.language_model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 94 |
+
"model.language_model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 95 |
+
"model.language_model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 96 |
+
"model.language_model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 97 |
+
"model.language_model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 98 |
+
"model.language_model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 99 |
+
"model.language_model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 100 |
+
"model.language_model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 101 |
+
"model.language_model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 102 |
+
"model.language_model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 103 |
+
"model.language_model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 104 |
+
"model.language_model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 105 |
+
"model.language_model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 106 |
+
"model.language_model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 107 |
+
"model.language_model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 108 |
+
"model.language_model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 109 |
+
"model.language_model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 110 |
+
"model.language_model.layers.17.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 111 |
+
"model.language_model.layers.17.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 112 |
+
"model.language_model.layers.17.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 113 |
+
"model.language_model.layers.17.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 114 |
+
"model.language_model.layers.17.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 115 |
+
"model.language_model.layers.17.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 116 |
+
"model.language_model.layers.17.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 117 |
+
"model.language_model.layers.17.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 118 |
+
"model.language_model.layers.17.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 119 |
+
"model.language_model.layers.17.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 120 |
+
"model.language_model.layers.18.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 121 |
+
"model.language_model.layers.18.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 122 |
+
"model.language_model.layers.18.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 123 |
+
"model.language_model.layers.18.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 124 |
+
"model.language_model.layers.18.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 125 |
+
"model.language_model.layers.18.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 126 |
+
"model.language_model.layers.18.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 127 |
+
"model.language_model.layers.18.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 128 |
+
"model.language_model.layers.18.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 129 |
+
"model.language_model.layers.18.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 130 |
+
"model.language_model.layers.18.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 131 |
+
"model.language_model.layers.19.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 132 |
+
"model.language_model.layers.19.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 133 |
+
"model.language_model.layers.19.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 134 |
+
"model.language_model.layers.19.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 135 |
+
"model.language_model.layers.19.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 136 |
+
"model.language_model.layers.19.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 137 |
+
"model.language_model.layers.19.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 138 |
+
"model.language_model.layers.19.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 139 |
+
"model.language_model.layers.19.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 140 |
+
"model.language_model.layers.19.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 141 |
+
"model.language_model.layers.19.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 142 |
+
"model.language_model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 143 |
+
"model.language_model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 144 |
+
"model.language_model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 145 |
+
"model.language_model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 146 |
+
"model.language_model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 147 |
+
"model.language_model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 148 |
+
"model.language_model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 149 |
+
"model.language_model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 150 |
+
"model.language_model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 151 |
+
"model.language_model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 152 |
+
"model.language_model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 153 |
+
"model.language_model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 154 |
+
"model.language_model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 155 |
+
"model.language_model.layers.20.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 156 |
+
"model.language_model.layers.20.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 157 |
+
"model.language_model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 158 |
+
"model.language_model.layers.20.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 159 |
+
"model.language_model.layers.20.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 160 |
+
"model.language_model.layers.20.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 161 |
+
"model.language_model.layers.20.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 162 |
+
"model.language_model.layers.20.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 163 |
+
"model.language_model.layers.20.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 164 |
+
"model.language_model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 165 |
+
"model.language_model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 166 |
+
"model.language_model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 167 |
+
"model.language_model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 168 |
+
"model.language_model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 169 |
+
"model.language_model.layers.21.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 170 |
+
"model.language_model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 171 |
+
"model.language_model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 172 |
+
"model.language_model.layers.21.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 173 |
+
"model.language_model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 174 |
+
"model.language_model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 175 |
+
"model.language_model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 176 |
+
"model.language_model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 177 |
+
"model.language_model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 178 |
+
"model.language_model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 179 |
+
"model.language_model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 180 |
+
"model.language_model.layers.22.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 181 |
+
"model.language_model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 182 |
+
"model.language_model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 183 |
+
"model.language_model.layers.22.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 184 |
+
"model.language_model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 185 |
+
"model.language_model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 186 |
+
"model.language_model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 187 |
+
"model.language_model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 188 |
+
"model.language_model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 189 |
+
"model.language_model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 190 |
+
"model.language_model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 191 |
+
"model.language_model.layers.23.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 192 |
+
"model.language_model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 193 |
+
"model.language_model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 194 |
+
"model.language_model.layers.23.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 195 |
+
"model.language_model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 196 |
+
"model.language_model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 197 |
+
"model.language_model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 198 |
+
"model.language_model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 199 |
+
"model.language_model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 200 |
+
"model.language_model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 201 |
+
"model.language_model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 202 |
+
"model.language_model.layers.24.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 203 |
+
"model.language_model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 204 |
+
"model.language_model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 205 |
+
"model.language_model.layers.24.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 206 |
+
"model.language_model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 207 |
+
"model.language_model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 208 |
+
"model.language_model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 209 |
+
"model.language_model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 210 |
+
"model.language_model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 211 |
+
"model.language_model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 212 |
+
"model.language_model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 213 |
+
"model.language_model.layers.25.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 214 |
+
"model.language_model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 215 |
+
"model.language_model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 216 |
+
"model.language_model.layers.25.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 217 |
+
"model.language_model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 218 |
+
"model.language_model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 219 |
+
"model.language_model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 220 |
+
"model.language_model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 221 |
+
"model.language_model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 222 |
+
"model.language_model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 223 |
+
"model.language_model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 224 |
+
"model.language_model.layers.26.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 225 |
+
"model.language_model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 226 |
+
"model.language_model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 227 |
+
"model.language_model.layers.26.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 228 |
+
"model.language_model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 229 |
+
"model.language_model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 230 |
+
"model.language_model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 231 |
+
"model.language_model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 232 |
+
"model.language_model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 233 |
+
"model.language_model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 234 |
+
"model.language_model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 235 |
+
"model.language_model.layers.27.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 236 |
+
"model.language_model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 237 |
+
"model.language_model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 238 |
+
"model.language_model.layers.27.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 239 |
+
"model.language_model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 240 |
+
"model.language_model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 241 |
+
"model.language_model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 242 |
+
"model.language_model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 243 |
+
"model.language_model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 244 |
+
"model.language_model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 245 |
+
"model.language_model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 246 |
+
"model.language_model.layers.28.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 247 |
+
"model.language_model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 248 |
+
"model.language_model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 249 |
+
"model.language_model.layers.28.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 250 |
+
"model.language_model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 251 |
+
"model.language_model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 252 |
+
"model.language_model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 253 |
+
"model.language_model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 254 |
+
"model.language_model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 255 |
+
"model.language_model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 256 |
+
"model.language_model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 257 |
+
"model.language_model.layers.29.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 258 |
+
"model.language_model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 259 |
+
"model.language_model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 260 |
+
"model.language_model.layers.29.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 261 |
+
"model.language_model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 262 |
+
"model.language_model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 263 |
+
"model.language_model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 264 |
+
"model.language_model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 265 |
+
"model.language_model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 266 |
+
"model.language_model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 267 |
+
"model.language_model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 268 |
+
"model.language_model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 269 |
+
"model.language_model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 270 |
+
"model.language_model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 271 |
+
"model.language_model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 272 |
+
"model.language_model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 273 |
+
"model.language_model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 274 |
+
"model.language_model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 275 |
+
"model.language_model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 276 |
+
"model.language_model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 277 |
+
"model.language_model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 278 |
+
"model.language_model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 279 |
+
"model.language_model.layers.30.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 280 |
+
"model.language_model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 281 |
+
"model.language_model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 282 |
+
"model.language_model.layers.30.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 283 |
+
"model.language_model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 284 |
+
"model.language_model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 285 |
+
"model.language_model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 286 |
+
"model.language_model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 287 |
+
"model.language_model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 288 |
+
"model.language_model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 289 |
+
"model.language_model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 290 |
+
"model.language_model.layers.31.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 291 |
+
"model.language_model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 292 |
+
"model.language_model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 293 |
+
"model.language_model.layers.31.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 294 |
+
"model.language_model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 295 |
+
"model.language_model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 296 |
+
"model.language_model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 297 |
+
"model.language_model.layers.32.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 298 |
+
"model.language_model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 299 |
+
"model.language_model.layers.32.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 300 |
+
"model.language_model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 301 |
+
"model.language_model.layers.32.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 302 |
+
"model.language_model.layers.32.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 303 |
+
"model.language_model.layers.32.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 304 |
+
"model.language_model.layers.32.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 305 |
+
"model.language_model.layers.32.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 306 |
+
"model.language_model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 307 |
+
"model.language_model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 308 |
+
"model.language_model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 309 |
+
"model.language_model.layers.33.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 310 |
+
"model.language_model.layers.33.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 311 |
+
"model.language_model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 312 |
+
"model.language_model.layers.33.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 313 |
+
"model.language_model.layers.33.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 314 |
+
"model.language_model.layers.33.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 315 |
+
"model.language_model.layers.33.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 316 |
+
"model.language_model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 317 |
+
"model.language_model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 318 |
+
"model.language_model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 319 |
+
"model.language_model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 320 |
+
"model.language_model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 321 |
+
"model.language_model.layers.34.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 322 |
+
"model.language_model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 323 |
+
"model.language_model.layers.34.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 324 |
+
"model.language_model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 325 |
+
"model.language_model.layers.34.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 326 |
+
"model.language_model.layers.34.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 327 |
+
"model.language_model.layers.34.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 328 |
+
"model.language_model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 329 |
+
"model.language_model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 330 |
+
"model.language_model.layers.35.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
| 331 |
+
"model.language_model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
| 332 |
+
"model.language_model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
| 333 |
+
"model.language_model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 334 |
+
"model.language_model.layers.35.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
| 335 |
+
"model.language_model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
| 336 |
+
"model.language_model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
| 337 |
+
"model.language_model.layers.35.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
| 338 |
+
"model.language_model.layers.35.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
| 339 |
+
"model.language_model.layers.35.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
| 340 |
+
"model.language_model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 341 |
+
"model.language_model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 342 |
+
"model.language_model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 343 |
+
"model.language_model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 344 |
+
"model.language_model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 345 |
+
"model.language_model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 346 |
+
"model.language_model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 347 |
+
"model.language_model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 348 |
+
"model.language_model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 349 |
+
"model.language_model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 350 |
+
"model.language_model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 351 |
+
"model.language_model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 352 |
+
"model.language_model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 353 |
+
"model.language_model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 354 |
+
"model.language_model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 355 |
+
"model.language_model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 356 |
+
"model.language_model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 357 |
+
"model.language_model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 358 |
+
"model.language_model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 359 |
+
"model.language_model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 360 |
+
"model.language_model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 361 |
+
"model.language_model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 362 |
+
"model.language_model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 363 |
+
"model.language_model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 364 |
+
"model.language_model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 365 |
+
"model.language_model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 366 |
+
"model.language_model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 367 |
+
"model.language_model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 368 |
+
"model.language_model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 369 |
+
"model.language_model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 370 |
+
"model.language_model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 371 |
+
"model.language_model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 372 |
+
"model.language_model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 373 |
+
"model.language_model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 374 |
+
"model.language_model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 375 |
+
"model.language_model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 376 |
+
"model.language_model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 377 |
+
"model.language_model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 378 |
+
"model.language_model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 379 |
+
"model.language_model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 380 |
+
"model.language_model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 381 |
+
"model.language_model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 382 |
+
"model.language_model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 383 |
+
"model.language_model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 384 |
+
"model.language_model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 385 |
+
"model.language_model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 386 |
+
"model.language_model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 387 |
+
"model.language_model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 388 |
+
"model.language_model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 389 |
+
"model.language_model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 390 |
+
"model.language_model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 391 |
+
"model.language_model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 392 |
+
"model.language_model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 393 |
+
"model.language_model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 394 |
+
"model.language_model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 395 |
+
"model.language_model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 396 |
+
"model.language_model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 397 |
+
"model.language_model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 398 |
+
"model.language_model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 399 |
+
"model.language_model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 400 |
+
"model.language_model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 401 |
+
"model.language_model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 402 |
+
"model.language_model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 403 |
+
"model.language_model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 404 |
+
"model.language_model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 405 |
+
"model.language_model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 406 |
+
"model.language_model.norm.weight": "model-00002-of-00002.safetensors",
|
| 407 |
+
"model.multi_modal_projector.linear_1.bias": "model-00002-of-00002.safetensors",
|
| 408 |
+
"model.multi_modal_projector.linear_1.weight": "model-00002-of-00002.safetensors",
|
| 409 |
+
"model.multi_modal_projector.linear_2.bias": "model-00002-of-00002.safetensors",
|
| 410 |
+
"model.multi_modal_projector.linear_2.weight": "model-00002-of-00002.safetensors",
|
| 411 |
+
"model.vision_tower.vision.blocks.0.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 412 |
+
"model.vision_tower.vision.blocks.0.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 413 |
+
"model.vision_tower.vision.blocks.0.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 414 |
+
"model.vision_tower.vision.blocks.0.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 415 |
+
"model.vision_tower.vision.blocks.0.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 416 |
+
"model.vision_tower.vision.blocks.0.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 417 |
+
"model.vision_tower.vision.blocks.0.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 418 |
+
"model.vision_tower.vision.blocks.0.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 419 |
+
"model.vision_tower.vision.blocks.0.norm1.bias": "model-00002-of-00002.safetensors",
|
| 420 |
+
"model.vision_tower.vision.blocks.0.norm1.weight": "model-00002-of-00002.safetensors",
|
| 421 |
+
"model.vision_tower.vision.blocks.0.norm2.bias": "model-00002-of-00002.safetensors",
|
| 422 |
+
"model.vision_tower.vision.blocks.0.norm2.weight": "model-00002-of-00002.safetensors",
|
| 423 |
+
"model.vision_tower.vision.blocks.1.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 424 |
+
"model.vision_tower.vision.blocks.1.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 425 |
+
"model.vision_tower.vision.blocks.1.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 426 |
+
"model.vision_tower.vision.blocks.1.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 427 |
+
"model.vision_tower.vision.blocks.1.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 428 |
+
"model.vision_tower.vision.blocks.1.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 429 |
+
"model.vision_tower.vision.blocks.1.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 430 |
+
"model.vision_tower.vision.blocks.1.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 431 |
+
"model.vision_tower.vision.blocks.1.norm1.bias": "model-00002-of-00002.safetensors",
|
| 432 |
+
"model.vision_tower.vision.blocks.1.norm1.weight": "model-00002-of-00002.safetensors",
|
| 433 |
+
"model.vision_tower.vision.blocks.1.norm2.bias": "model-00002-of-00002.safetensors",
|
| 434 |
+
"model.vision_tower.vision.blocks.1.norm2.weight": "model-00002-of-00002.safetensors",
|
| 435 |
+
"model.vision_tower.vision.blocks.10.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 436 |
+
"model.vision_tower.vision.blocks.10.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 437 |
+
"model.vision_tower.vision.blocks.10.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 438 |
+
"model.vision_tower.vision.blocks.10.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 439 |
+
"model.vision_tower.vision.blocks.10.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 440 |
+
"model.vision_tower.vision.blocks.10.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 441 |
+
"model.vision_tower.vision.blocks.10.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 442 |
+
"model.vision_tower.vision.blocks.10.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 443 |
+
"model.vision_tower.vision.blocks.10.norm1.bias": "model-00002-of-00002.safetensors",
|
| 444 |
+
"model.vision_tower.vision.blocks.10.norm1.weight": "model-00002-of-00002.safetensors",
|
| 445 |
+
"model.vision_tower.vision.blocks.10.norm2.bias": "model-00002-of-00002.safetensors",
|
| 446 |
+
"model.vision_tower.vision.blocks.10.norm2.weight": "model-00002-of-00002.safetensors",
|
| 447 |
+
"model.vision_tower.vision.blocks.11.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 448 |
+
"model.vision_tower.vision.blocks.11.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 449 |
+
"model.vision_tower.vision.blocks.11.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 450 |
+
"model.vision_tower.vision.blocks.11.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 451 |
+
"model.vision_tower.vision.blocks.11.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 452 |
+
"model.vision_tower.vision.blocks.11.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 453 |
+
"model.vision_tower.vision.blocks.11.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 454 |
+
"model.vision_tower.vision.blocks.11.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 455 |
+
"model.vision_tower.vision.blocks.11.norm1.bias": "model-00002-of-00002.safetensors",
|
| 456 |
+
"model.vision_tower.vision.blocks.11.norm1.weight": "model-00002-of-00002.safetensors",
|
| 457 |
+
"model.vision_tower.vision.blocks.11.norm2.bias": "model-00002-of-00002.safetensors",
|
| 458 |
+
"model.vision_tower.vision.blocks.11.norm2.weight": "model-00002-of-00002.safetensors",
|
| 459 |
+
"model.vision_tower.vision.blocks.12.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 460 |
+
"model.vision_tower.vision.blocks.12.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 461 |
+
"model.vision_tower.vision.blocks.12.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 462 |
+
"model.vision_tower.vision.blocks.12.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 463 |
+
"model.vision_tower.vision.blocks.12.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 464 |
+
"model.vision_tower.vision.blocks.12.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 465 |
+
"model.vision_tower.vision.blocks.12.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 466 |
+
"model.vision_tower.vision.blocks.12.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 467 |
+
"model.vision_tower.vision.blocks.12.norm1.bias": "model-00002-of-00002.safetensors",
|
| 468 |
+
"model.vision_tower.vision.blocks.12.norm1.weight": "model-00002-of-00002.safetensors",
|
| 469 |
+
"model.vision_tower.vision.blocks.12.norm2.bias": "model-00002-of-00002.safetensors",
|
| 470 |
+
"model.vision_tower.vision.blocks.12.norm2.weight": "model-00002-of-00002.safetensors",
|
| 471 |
+
"model.vision_tower.vision.blocks.13.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 472 |
+
"model.vision_tower.vision.blocks.13.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 473 |
+
"model.vision_tower.vision.blocks.13.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 474 |
+
"model.vision_tower.vision.blocks.13.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 475 |
+
"model.vision_tower.vision.blocks.13.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 476 |
+
"model.vision_tower.vision.blocks.13.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 477 |
+
"model.vision_tower.vision.blocks.13.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 478 |
+
"model.vision_tower.vision.blocks.13.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 479 |
+
"model.vision_tower.vision.blocks.13.norm1.bias": "model-00002-of-00002.safetensors",
|
| 480 |
+
"model.vision_tower.vision.blocks.13.norm1.weight": "model-00002-of-00002.safetensors",
|
| 481 |
+
"model.vision_tower.vision.blocks.13.norm2.bias": "model-00002-of-00002.safetensors",
|
| 482 |
+
"model.vision_tower.vision.blocks.13.norm2.weight": "model-00002-of-00002.safetensors",
|
| 483 |
+
"model.vision_tower.vision.blocks.14.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 484 |
+
"model.vision_tower.vision.blocks.14.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 485 |
+
"model.vision_tower.vision.blocks.14.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 486 |
+
"model.vision_tower.vision.blocks.14.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 487 |
+
"model.vision_tower.vision.blocks.14.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 488 |
+
"model.vision_tower.vision.blocks.14.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 489 |
+
"model.vision_tower.vision.blocks.14.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 490 |
+
"model.vision_tower.vision.blocks.14.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 491 |
+
"model.vision_tower.vision.blocks.14.norm1.bias": "model-00002-of-00002.safetensors",
|
| 492 |
+
"model.vision_tower.vision.blocks.14.norm1.weight": "model-00002-of-00002.safetensors",
|
| 493 |
+
"model.vision_tower.vision.blocks.14.norm2.bias": "model-00002-of-00002.safetensors",
|
| 494 |
+
"model.vision_tower.vision.blocks.14.norm2.weight": "model-00002-of-00002.safetensors",
|
| 495 |
+
"model.vision_tower.vision.blocks.15.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 496 |
+
"model.vision_tower.vision.blocks.15.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 497 |
+
"model.vision_tower.vision.blocks.15.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 498 |
+
"model.vision_tower.vision.blocks.15.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 499 |
+
"model.vision_tower.vision.blocks.15.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 500 |
+
"model.vision_tower.vision.blocks.15.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 501 |
+
"model.vision_tower.vision.blocks.15.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 502 |
+
"model.vision_tower.vision.blocks.15.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 503 |
+
"model.vision_tower.vision.blocks.15.norm1.bias": "model-00002-of-00002.safetensors",
|
| 504 |
+
"model.vision_tower.vision.blocks.15.norm1.weight": "model-00002-of-00002.safetensors",
|
| 505 |
+
"model.vision_tower.vision.blocks.15.norm2.bias": "model-00002-of-00002.safetensors",
|
| 506 |
+
"model.vision_tower.vision.blocks.15.norm2.weight": "model-00002-of-00002.safetensors",
|
| 507 |
+
"model.vision_tower.vision.blocks.16.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 508 |
+
"model.vision_tower.vision.blocks.16.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 509 |
+
"model.vision_tower.vision.blocks.16.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 510 |
+
"model.vision_tower.vision.blocks.16.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 511 |
+
"model.vision_tower.vision.blocks.16.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 512 |
+
"model.vision_tower.vision.blocks.16.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 513 |
+
"model.vision_tower.vision.blocks.16.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 514 |
+
"model.vision_tower.vision.blocks.16.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 515 |
+
"model.vision_tower.vision.blocks.16.norm1.bias": "model-00002-of-00002.safetensors",
|
| 516 |
+
"model.vision_tower.vision.blocks.16.norm1.weight": "model-00002-of-00002.safetensors",
|
| 517 |
+
"model.vision_tower.vision.blocks.16.norm2.bias": "model-00002-of-00002.safetensors",
|
| 518 |
+
"model.vision_tower.vision.blocks.16.norm2.weight": "model-00002-of-00002.safetensors",
|
| 519 |
+
"model.vision_tower.vision.blocks.17.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 520 |
+
"model.vision_tower.vision.blocks.17.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 521 |
+
"model.vision_tower.vision.blocks.17.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 522 |
+
"model.vision_tower.vision.blocks.17.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 523 |
+
"model.vision_tower.vision.blocks.17.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 524 |
+
"model.vision_tower.vision.blocks.17.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 525 |
+
"model.vision_tower.vision.blocks.17.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 526 |
+
"model.vision_tower.vision.blocks.17.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 527 |
+
"model.vision_tower.vision.blocks.17.norm1.bias": "model-00002-of-00002.safetensors",
|
| 528 |
+
"model.vision_tower.vision.blocks.17.norm1.weight": "model-00002-of-00002.safetensors",
|
| 529 |
+
"model.vision_tower.vision.blocks.17.norm2.bias": "model-00002-of-00002.safetensors",
|
| 530 |
+
"model.vision_tower.vision.blocks.17.norm2.weight": "model-00002-of-00002.safetensors",
|
| 531 |
+
"model.vision_tower.vision.blocks.18.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 532 |
+
"model.vision_tower.vision.blocks.18.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 533 |
+
"model.vision_tower.vision.blocks.18.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 534 |
+
"model.vision_tower.vision.blocks.18.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 535 |
+
"model.vision_tower.vision.blocks.18.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 536 |
+
"model.vision_tower.vision.blocks.18.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 537 |
+
"model.vision_tower.vision.blocks.18.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 538 |
+
"model.vision_tower.vision.blocks.18.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 539 |
+
"model.vision_tower.vision.blocks.18.norm1.bias": "model-00002-of-00002.safetensors",
|
| 540 |
+
"model.vision_tower.vision.blocks.18.norm1.weight": "model-00002-of-00002.safetensors",
|
| 541 |
+
"model.vision_tower.vision.blocks.18.norm2.bias": "model-00002-of-00002.safetensors",
|
| 542 |
+
"model.vision_tower.vision.blocks.18.norm2.weight": "model-00002-of-00002.safetensors",
|
| 543 |
+
"model.vision_tower.vision.blocks.19.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 544 |
+
"model.vision_tower.vision.blocks.19.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 545 |
+
"model.vision_tower.vision.blocks.19.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 546 |
+
"model.vision_tower.vision.blocks.19.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 547 |
+
"model.vision_tower.vision.blocks.19.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 548 |
+
"model.vision_tower.vision.blocks.19.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 549 |
+
"model.vision_tower.vision.blocks.19.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 550 |
+
"model.vision_tower.vision.blocks.19.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 551 |
+
"model.vision_tower.vision.blocks.19.norm1.bias": "model-00002-of-00002.safetensors",
|
| 552 |
+
"model.vision_tower.vision.blocks.19.norm1.weight": "model-00002-of-00002.safetensors",
|
| 553 |
+
"model.vision_tower.vision.blocks.19.norm2.bias": "model-00002-of-00002.safetensors",
|
| 554 |
+
"model.vision_tower.vision.blocks.19.norm2.weight": "model-00002-of-00002.safetensors",
|
| 555 |
+
"model.vision_tower.vision.blocks.2.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 556 |
+
"model.vision_tower.vision.blocks.2.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 557 |
+
"model.vision_tower.vision.blocks.2.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 558 |
+
"model.vision_tower.vision.blocks.2.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 559 |
+
"model.vision_tower.vision.blocks.2.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 560 |
+
"model.vision_tower.vision.blocks.2.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 561 |
+
"model.vision_tower.vision.blocks.2.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 562 |
+
"model.vision_tower.vision.blocks.2.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 563 |
+
"model.vision_tower.vision.blocks.2.norm1.bias": "model-00002-of-00002.safetensors",
|
| 564 |
+
"model.vision_tower.vision.blocks.2.norm1.weight": "model-00002-of-00002.safetensors",
|
| 565 |
+
"model.vision_tower.vision.blocks.2.norm2.bias": "model-00002-of-00002.safetensors",
|
| 566 |
+
"model.vision_tower.vision.blocks.2.norm2.weight": "model-00002-of-00002.safetensors",
|
| 567 |
+
"model.vision_tower.vision.blocks.20.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 568 |
+
"model.vision_tower.vision.blocks.20.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 569 |
+
"model.vision_tower.vision.blocks.20.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 570 |
+
"model.vision_tower.vision.blocks.20.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 571 |
+
"model.vision_tower.vision.blocks.20.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 572 |
+
"model.vision_tower.vision.blocks.20.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 573 |
+
"model.vision_tower.vision.blocks.20.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 574 |
+
"model.vision_tower.vision.blocks.20.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 575 |
+
"model.vision_tower.vision.blocks.20.norm1.bias": "model-00002-of-00002.safetensors",
|
| 576 |
+
"model.vision_tower.vision.blocks.20.norm1.weight": "model-00002-of-00002.safetensors",
|
| 577 |
+
"model.vision_tower.vision.blocks.20.norm2.bias": "model-00002-of-00002.safetensors",
|
| 578 |
+
"model.vision_tower.vision.blocks.20.norm2.weight": "model-00002-of-00002.safetensors",
|
| 579 |
+
"model.vision_tower.vision.blocks.21.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 580 |
+
"model.vision_tower.vision.blocks.21.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 581 |
+
"model.vision_tower.vision.blocks.21.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 582 |
+
"model.vision_tower.vision.blocks.21.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 583 |
+
"model.vision_tower.vision.blocks.21.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 584 |
+
"model.vision_tower.vision.blocks.21.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 585 |
+
"model.vision_tower.vision.blocks.21.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 586 |
+
"model.vision_tower.vision.blocks.21.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 587 |
+
"model.vision_tower.vision.blocks.21.norm1.bias": "model-00002-of-00002.safetensors",
|
| 588 |
+
"model.vision_tower.vision.blocks.21.norm1.weight": "model-00002-of-00002.safetensors",
|
| 589 |
+
"model.vision_tower.vision.blocks.21.norm2.bias": "model-00002-of-00002.safetensors",
|
| 590 |
+
"model.vision_tower.vision.blocks.21.norm2.weight": "model-00002-of-00002.safetensors",
|
| 591 |
+
"model.vision_tower.vision.blocks.22.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 592 |
+
"model.vision_tower.vision.blocks.22.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 593 |
+
"model.vision_tower.vision.blocks.22.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 594 |
+
"model.vision_tower.vision.blocks.22.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 595 |
+
"model.vision_tower.vision.blocks.22.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 596 |
+
"model.vision_tower.vision.blocks.22.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 597 |
+
"model.vision_tower.vision.blocks.22.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 598 |
+
"model.vision_tower.vision.blocks.22.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 599 |
+
"model.vision_tower.vision.blocks.22.norm1.bias": "model-00002-of-00002.safetensors",
|
| 600 |
+
"model.vision_tower.vision.blocks.22.norm1.weight": "model-00002-of-00002.safetensors",
|
| 601 |
+
"model.vision_tower.vision.blocks.22.norm2.bias": "model-00002-of-00002.safetensors",
|
| 602 |
+
"model.vision_tower.vision.blocks.22.norm2.weight": "model-00002-of-00002.safetensors",
|
| 603 |
+
"model.vision_tower.vision.blocks.23.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 604 |
+
"model.vision_tower.vision.blocks.23.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 605 |
+
"model.vision_tower.vision.blocks.23.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 606 |
+
"model.vision_tower.vision.blocks.23.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 607 |
+
"model.vision_tower.vision.blocks.23.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 608 |
+
"model.vision_tower.vision.blocks.23.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 609 |
+
"model.vision_tower.vision.blocks.23.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 610 |
+
"model.vision_tower.vision.blocks.23.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 611 |
+
"model.vision_tower.vision.blocks.23.norm1.bias": "model-00002-of-00002.safetensors",
|
| 612 |
+
"model.vision_tower.vision.blocks.23.norm1.weight": "model-00002-of-00002.safetensors",
|
| 613 |
+
"model.vision_tower.vision.blocks.23.norm2.bias": "model-00002-of-00002.safetensors",
|
| 614 |
+
"model.vision_tower.vision.blocks.23.norm2.weight": "model-00002-of-00002.safetensors",
|
| 615 |
+
"model.vision_tower.vision.blocks.3.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 616 |
+
"model.vision_tower.vision.blocks.3.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 617 |
+
"model.vision_tower.vision.blocks.3.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 618 |
+
"model.vision_tower.vision.blocks.3.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 619 |
+
"model.vision_tower.vision.blocks.3.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 620 |
+
"model.vision_tower.vision.blocks.3.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 621 |
+
"model.vision_tower.vision.blocks.3.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 622 |
+
"model.vision_tower.vision.blocks.3.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 623 |
+
"model.vision_tower.vision.blocks.3.norm1.bias": "model-00002-of-00002.safetensors",
|
| 624 |
+
"model.vision_tower.vision.blocks.3.norm1.weight": "model-00002-of-00002.safetensors",
|
| 625 |
+
"model.vision_tower.vision.blocks.3.norm2.bias": "model-00002-of-00002.safetensors",
|
| 626 |
+
"model.vision_tower.vision.blocks.3.norm2.weight": "model-00002-of-00002.safetensors",
|
| 627 |
+
"model.vision_tower.vision.blocks.4.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 628 |
+
"model.vision_tower.vision.blocks.4.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 629 |
+
"model.vision_tower.vision.blocks.4.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 630 |
+
"model.vision_tower.vision.blocks.4.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 631 |
+
"model.vision_tower.vision.blocks.4.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 632 |
+
"model.vision_tower.vision.blocks.4.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 633 |
+
"model.vision_tower.vision.blocks.4.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 634 |
+
"model.vision_tower.vision.blocks.4.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 635 |
+
"model.vision_tower.vision.blocks.4.norm1.bias": "model-00002-of-00002.safetensors",
|
| 636 |
+
"model.vision_tower.vision.blocks.4.norm1.weight": "model-00002-of-00002.safetensors",
|
| 637 |
+
"model.vision_tower.vision.blocks.4.norm2.bias": "model-00002-of-00002.safetensors",
|
| 638 |
+
"model.vision_tower.vision.blocks.4.norm2.weight": "model-00002-of-00002.safetensors",
|
| 639 |
+
"model.vision_tower.vision.blocks.5.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 640 |
+
"model.vision_tower.vision.blocks.5.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 641 |
+
"model.vision_tower.vision.blocks.5.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 642 |
+
"model.vision_tower.vision.blocks.5.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 643 |
+
"model.vision_tower.vision.blocks.5.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 644 |
+
"model.vision_tower.vision.blocks.5.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 645 |
+
"model.vision_tower.vision.blocks.5.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 646 |
+
"model.vision_tower.vision.blocks.5.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 647 |
+
"model.vision_tower.vision.blocks.5.norm1.bias": "model-00002-of-00002.safetensors",
|
| 648 |
+
"model.vision_tower.vision.blocks.5.norm1.weight": "model-00002-of-00002.safetensors",
|
| 649 |
+
"model.vision_tower.vision.blocks.5.norm2.bias": "model-00002-of-00002.safetensors",
|
| 650 |
+
"model.vision_tower.vision.blocks.5.norm2.weight": "model-00002-of-00002.safetensors",
|
| 651 |
+
"model.vision_tower.vision.blocks.6.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 652 |
+
"model.vision_tower.vision.blocks.6.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 653 |
+
"model.vision_tower.vision.blocks.6.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 654 |
+
"model.vision_tower.vision.blocks.6.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 655 |
+
"model.vision_tower.vision.blocks.6.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 656 |
+
"model.vision_tower.vision.blocks.6.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 657 |
+
"model.vision_tower.vision.blocks.6.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 658 |
+
"model.vision_tower.vision.blocks.6.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 659 |
+
"model.vision_tower.vision.blocks.6.norm1.bias": "model-00002-of-00002.safetensors",
|
| 660 |
+
"model.vision_tower.vision.blocks.6.norm1.weight": "model-00002-of-00002.safetensors",
|
| 661 |
+
"model.vision_tower.vision.blocks.6.norm2.bias": "model-00002-of-00002.safetensors",
|
| 662 |
+
"model.vision_tower.vision.blocks.6.norm2.weight": "model-00002-of-00002.safetensors",
|
| 663 |
+
"model.vision_tower.vision.blocks.7.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 664 |
+
"model.vision_tower.vision.blocks.7.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 665 |
+
"model.vision_tower.vision.blocks.7.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 666 |
+
"model.vision_tower.vision.blocks.7.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 667 |
+
"model.vision_tower.vision.blocks.7.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 668 |
+
"model.vision_tower.vision.blocks.7.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 669 |
+
"model.vision_tower.vision.blocks.7.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 670 |
+
"model.vision_tower.vision.blocks.7.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 671 |
+
"model.vision_tower.vision.blocks.7.norm1.bias": "model-00002-of-00002.safetensors",
|
| 672 |
+
"model.vision_tower.vision.blocks.7.norm1.weight": "model-00002-of-00002.safetensors",
|
| 673 |
+
"model.vision_tower.vision.blocks.7.norm2.bias": "model-00002-of-00002.safetensors",
|
| 674 |
+
"model.vision_tower.vision.blocks.7.norm2.weight": "model-00002-of-00002.safetensors",
|
| 675 |
+
"model.vision_tower.vision.blocks.8.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 676 |
+
"model.vision_tower.vision.blocks.8.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 677 |
+
"model.vision_tower.vision.blocks.8.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 678 |
+
"model.vision_tower.vision.blocks.8.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 679 |
+
"model.vision_tower.vision.blocks.8.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 680 |
+
"model.vision_tower.vision.blocks.8.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 681 |
+
"model.vision_tower.vision.blocks.8.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 682 |
+
"model.vision_tower.vision.blocks.8.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 683 |
+
"model.vision_tower.vision.blocks.8.norm1.bias": "model-00002-of-00002.safetensors",
|
| 684 |
+
"model.vision_tower.vision.blocks.8.norm1.weight": "model-00002-of-00002.safetensors",
|
| 685 |
+
"model.vision_tower.vision.blocks.8.norm2.bias": "model-00002-of-00002.safetensors",
|
| 686 |
+
"model.vision_tower.vision.blocks.8.norm2.weight": "model-00002-of-00002.safetensors",
|
| 687 |
+
"model.vision_tower.vision.blocks.9.attn.proj.bias": "model-00002-of-00002.safetensors",
|
| 688 |
+
"model.vision_tower.vision.blocks.9.attn.proj.weight": "model-00002-of-00002.safetensors",
|
| 689 |
+
"model.vision_tower.vision.blocks.9.attn.qkv.bias": "model-00002-of-00002.safetensors",
|
| 690 |
+
"model.vision_tower.vision.blocks.9.attn.qkv.weight": "model-00002-of-00002.safetensors",
|
| 691 |
+
"model.vision_tower.vision.blocks.9.mlp.linear_fc1.bias": "model-00002-of-00002.safetensors",
|
| 692 |
+
"model.vision_tower.vision.blocks.9.mlp.linear_fc1.weight": "model-00002-of-00002.safetensors",
|
| 693 |
+
"model.vision_tower.vision.blocks.9.mlp.linear_fc2.bias": "model-00002-of-00002.safetensors",
|
| 694 |
+
"model.vision_tower.vision.blocks.9.mlp.linear_fc2.weight": "model-00002-of-00002.safetensors",
|
| 695 |
+
"model.vision_tower.vision.blocks.9.norm1.bias": "model-00002-of-00002.safetensors",
|
| 696 |
+
"model.vision_tower.vision.blocks.9.norm1.weight": "model-00002-of-00002.safetensors",
|
| 697 |
+
"model.vision_tower.vision.blocks.9.norm2.bias": "model-00002-of-00002.safetensors",
|
| 698 |
+
"model.vision_tower.vision.blocks.9.norm2.weight": "model-00002-of-00002.safetensors",
|
| 699 |
+
"model.vision_tower.vision.final_layernorm.bias": "model-00002-of-00002.safetensors",
|
| 700 |
+
"model.vision_tower.vision.final_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 701 |
+
"model.vision_tower.vision.patch_embed.proj.bias": "model-00002-of-00002.safetensors",
|
| 702 |
+
"model.vision_tower.vision.patch_embed.proj.weight": "model-00002-of-00002.safetensors",
|
| 703 |
+
"model.vision_tower.vision.pos_embed.weight": "model-00002-of-00002.safetensors"
|
| 704 |
+
}
|
| 705 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/modeling_qwen3_5vit_qwen3.py
ADDED
|
@@ -0,0 +1,351 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
modeling_qwen3_5vit_qwen3.py — Qwen3.5 Vision as SigLIP-compat vision_tower in LlavaOnevision.
|
| 3 |
+
|
| 4 |
+
设计：与 `modeling_qwen3vlvit_qwen3.py` 严格同构，仅换 vision backbone 源：
|
| 5 |
+
- Qwen3VLVisionModel → Qwen3_5VisionModel（继承关系：Qwen3_5VisionModel(Qwen3VLVisionModel) 去 DeepStack）
|
| 6 |
+
- Qwen3VLVisionConfig → Qwen3_5VisionConfig（父类用 AttributeError 哨兵屏蔽 deepstack_visual_indexes）
|
| 7 |
+
|
| 8 |
+
其余（Adapter 契约翻译、MLP projector + pre_norm、LlavaOnevision 继承 wire class）与
|
| 9 |
+
Qwen3-VL ViT pipeline 完全一致。两条 pipeline 并存意义：DeepStack ablation 天然实验组。
|
| 10 |
+
|
| 11 |
+
类层级：
|
| 12 |
+
Qwen3_5ViTBackbone(Qwen3_5VisionModel) — 去 merger，保持 NaViT 契约
|
| 13 |
+
Qwen3_5ViTAsSiglipAdapter(nn.Module) — 持有 Backbone，做 SigLIP ↔ NaViT 契约翻译
|
| 14 |
+
|
| 15 |
+
三方对比公平性：定 384×384 AnyRes tile + 同款 projector 骨架 + 同款 Qwen3-1.7B LLM。
|
| 16 |
+
"""
|
| 17 |
+
|
| 18 |
+
import math
|
| 19 |
+
import os
|
| 20 |
+
import sys
|
| 21 |
+
from typing import Optional
|
| 22 |
+
|
| 23 |
+
import torch
|
| 24 |
+
import torch.nn as nn
|
| 25 |
+
import torch.nn.functional as F
|
| 26 |
+
from transformers import (
|
| 27 |
+
AutoConfig,
|
| 28 |
+
AutoModel,
|
| 29 |
+
AutoModelForCausalLM,
|
| 30 |
+
LlavaOnevisionConfig,
|
| 31 |
+
LlavaOnevisionForConditionalGeneration,
|
| 32 |
+
LlavaOnevisionModel,
|
| 33 |
+
LlavaOnevisionPreTrainedModel,
|
| 34 |
+
Qwen3Config,
|
| 35 |
+
)
|
| 36 |
+
from transformers.activations import ACT2FN
|
| 37 |
+
from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
|
| 38 |
+
from transformers.models.qwen3_5.configuration_qwen3_5 import Qwen3_5VisionConfig
|
| 39 |
+
from transformers.models.qwen3_5.modeling_qwen3_5 import Qwen3_5VisionModel
|
| 40 |
+
|
| 41 |
+
# Shared layout-permutation utility lives in declip_qwenvit (single source of
|
| 42 |
+
# truth — same code path runs in declip-training-side qk_cosine reorder).
|
| 43 |
+
# Add VisionEncoder repo root to sys.path so this modeling file is importable
|
| 44 |
+
# even when the package isn't pip-installed (ms-swift integration loads it
|
| 45 |
+
# via dynamic plugin path).
|
| 46 |
+
_REPO_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
|
| 47 |
+
if _REPO_ROOT not in sys.path:
|
| 48 |
+
sys.path.append(_REPO_ROOT)
|
| 49 |
+
from declip_qwenvit.model.qwen3vit_qk import block_merge_to_row_major_permutation # noqa: E402
|
| 50 |
+
|
| 51 |
+
|
| 52 |
+
class LlavaQwen3_5ViTConfig(LlavaOnevisionConfig):
|
| 53 |
+
"""LlavaOnevisionConfig 子类，vision_config 类型换成 Qwen3_5VisionConfig。
|
| 54 |
+
|
| 55 |
+
与 Qwen3-VL ViT 版的差异：
|
| 56 |
+
- sub_configs.vision_config 用 Qwen3_5VisionConfig
|
| 57 |
+
- 不再设置 deepstack_visual_indexes（Qwen3_5VisionConfig 用 AttributeError 哨兵屏蔽此字段）
|
| 58 |
+
|
| 59 |
+
其余同 LlavaQwen3VLViTConfig（tile_size 默认 384；vision_feature_select_strategy='full'
|
| 60 |
+
必须 override，Qwen3.5 ViT 无 CLS token）。
|
| 61 |
+
"""
|
| 62 |
+
|
| 63 |
+
model_type = "llava_qwen3_5vit_qwen3"
|
| 64 |
+
sub_configs = {"vision_config": Qwen3_5VisionConfig, "text_config": Qwen3Config}
|
| 65 |
+
|
| 66 |
+
def __init__(
|
| 67 |
+
self,
|
| 68 |
+
vision_config=None,
|
| 69 |
+
text_config=None,
|
| 70 |
+
tile_size: int = 384,
|
| 71 |
+
**kwargs,
|
| 72 |
+
):
|
| 73 |
+
if isinstance(vision_config, dict):
|
| 74 |
+
vision_config = Qwen3_5VisionConfig(**vision_config)
|
| 75 |
+
elif vision_config is None:
|
| 76 |
+
vision_config = Qwen3_5VisionConfig()
|
| 77 |
+
# WHY 无 `vision_config.deepstack_visual_indexes = []`（对比 Qwen3-VL ViT 版）：
|
| 78 |
+
# Qwen3_5VisionConfig 父类用 AttributeError() 哨兵显式屏蔽此字段，设置会报 AttributeError
|
| 79 |
+
# LlavaOnevision.pack_image_features 用这个作为 tile 像素大小（不是 patch_size）
|
| 80 |
+
vision_config.image_size = tile_size
|
| 81 |
+
|
| 82 |
+
if isinstance(text_config, dict):
|
| 83 |
+
text_config = Qwen3Config(**text_config)
|
| 84 |
+
elif text_config is None:
|
| 85 |
+
text_config = Qwen3Config()
|
| 86 |
+
|
| 87 |
+
# 父类默认 select_strategy='default' 会跳首 token (CLS) — Qwen3.5 ViT 无 CLS 必须用 'full'
|
| 88 |
+
kwargs.setdefault("vision_feature_select_strategy", "full")
|
| 89 |
+
super().__init__(vision_config=vision_config, text_config=text_config, **kwargs)
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
class Qwen3_5ViTBackbone(Qwen3_5VisionModel):
|
| 93 |
+
"""Qwen3.5 Vision 去除原生 patch_merger 的 backbone 版本（V6 final_layernorm fix (2026-05-16): append final_layernorm）。
|
| 94 |
+
|
| 95 |
+
构造时把 merger.norm 的预训练权重抠到 final_layernorm，然后 `del self.merger`
|
| 96 |
+
释放 ~37M 参数（保留 norm 的 LN 焊到末端做 post_layernorm 角色，丢弃 spatial
|
| 97 |
+
shuffle + linear_fc1/fc2，那对应 LlavaOV projector 的职责）。
|
| 98 |
+
|
| 99 |
+
架构对称（V6 final_layernorm fix (2026-05-16) 修复）：
|
| 100 |
+
SigLIP2: encoder → post_layernorm → last_hidden_state → LlavaOV MLP → LLM
|
| 101 |
+
V6 final_layernorm fix (2026-05-16): encoder → final_layernorm → last_hidden_state → LlavaOV MLP → LLM
|
| 102 |
+
|
| 103 |
+
forward 跑完 transformer blocks 后过 final_layernorm，再返回。下游 LlavaOnevision
|
| 104 |
+
pack_image_features 的 AnyRes 2×2 pool 接管原 merger 的空间合并职责。
|
| 105 |
+
|
| 106 |
+
输入输出契约与父类 Qwen3_5VisionModel 一致（NaViT flat）：
|
| 107 |
+
forward(hidden_states=[L, patch_dim], grid_thw=[N, 3])
|
| 108 |
+
→ BaseModelOutput(last_hidden_state=[L, hidden_size])
|
| 109 |
+
|
| 110 |
+
forward 主体 1:1 对照 `Qwen3_5VisionModel.forward`（已无 deepstack loop，比
|
| 111 |
+
Qwen3VLVisionModel.forward 更短），仅跳过末尾 `self.merger(x)`，改为 final_layernorm。
|
| 112 |
+
"""
|
| 113 |
+
|
| 114 |
+
def __init__(self, config):
|
| 115 |
+
super().__init__(config)
|
| 116 |
+
# V6 final_layernorm fix (2026-05-16): extract merger.norm pretrained weights into final_layernorm.
|
| 117 |
+
# Default init: even if ckpt 阶段没 inject final_layernorm.* (e.g. stock
|
| 118 |
+
# bootstrap path that filters merger.*), final_layernorm 仍持有 merger.norm
|
| 119 |
+
# 的预训练值, 不是 random — 这是防止 silent corruption 的兜底.
|
| 120 |
+
ln_w = self.merger.norm.weight.detach().clone()
|
| 121 |
+
ln_b = self.merger.norm.bias.detach().clone()
|
| 122 |
+
del self.merger
|
| 123 |
+
self.final_layernorm = nn.LayerNorm(config.hidden_size, eps=1e-6)
|
| 124 |
+
self.final_layernorm.weight.data.copy_(ln_w)
|
| 125 |
+
self.final_layernorm.bias.data.copy_(ln_b)
|
| 126 |
+
|
| 127 |
+
def forward(self, hidden_states, grid_thw, **kwargs):
|
| 128 |
+
hidden_states = self.patch_embed(hidden_states)
|
| 129 |
+
|
| 130 |
+
pos_embeds = self.fast_pos_embed_interpolate(grid_thw)
|
| 131 |
+
hidden_states = hidden_states + pos_embeds
|
| 132 |
+
|
| 133 |
+
rotary_pos_emb = self.rot_pos_emb(grid_thw)
|
| 134 |
+
seq_len, _ = hidden_states.size()
|
| 135 |
+
rotary_pos_emb = rotary_pos_emb.reshape(seq_len, -1)
|
| 136 |
+
emb = torch.cat((rotary_pos_emb, rotary_pos_emb), dim=-1)
|
| 137 |
+
position_embeddings = (emb.cos(), emb.sin())
|
| 138 |
+
|
| 139 |
+
cu_seqlens = torch.repeat_interleave(
|
| 140 |
+
grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]
|
| 141 |
+
).cumsum(dim=0, dtype=torch.int32)
|
| 142 |
+
cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
|
| 143 |
+
|
| 144 |
+
for blk in self.blocks:
|
| 145 |
+
hidden_states = blk(
|
| 146 |
+
hidden_states,
|
| 147 |
+
cu_seqlens=cu_seqlens,
|
| 148 |
+
position_embeddings=position_embeddings,
|
| 149 |
+
**kwargs,
|
| 150 |
+
)
|
| 151 |
+
|
| 152 |
+
# V6 final_layernorm fix (2026-05-16): appended final LayerNorm — mirrors SigLIP2's post_layernorm.
|
| 153 |
+
# Per-token affine; layout-invariant (reorder happens in adapter).
|
| 154 |
+
hidden_states = self.final_layernorm(hidden_states)
|
| 155 |
+
|
| 156 |
+
return BaseModelOutput(last_hidden_state=hidden_states)
|
| 157 |
+
|
| 158 |
+
|
| 159 |
+
class Qwen3_5ViTAsSiglipAdapter(nn.Module):
|
| 160 |
+
"""SigLIP 契约 → NaViT 契约的翻译层。持有 Qwen3_5ViTBackbone。
|
| 161 |
+
|
| 162 |
+
对外暴露 SigLIP 式 forward(pixel_values=[N,3,H,W]) → BaseModelOutputWithPooling，
|
| 163 |
+
供 LlavaOnevision 消费；对内按官方 _preprocess 的 reshape 链把 pixel_values
|
| 164 |
+
转成 NaViT flat + grid_thw 喂给 Backbone。
|
| 165 |
+
|
| 166 |
+
reshape 链 1:1 照抄 transformers 官方 Qwen2VLImageProcessorFast._preprocess
|
| 167 |
+
(video_processing_qwen3_vl.py L227-252) —— Qwen3.5 无独立 image_processor，复用 Qwen3-VL 格式。
|
| 168 |
+
"""
|
| 169 |
+
|
| 170 |
+
def __init__(self, vision_config: Qwen3_5VisionConfig):
|
| 171 |
+
super().__init__()
|
| 172 |
+
self.vision = Qwen3_5ViTBackbone(vision_config)
|
| 173 |
+
self.config = vision_config
|
| 174 |
+
|
| 175 |
+
@property
|
| 176 |
+
def dtype(self):
|
| 177 |
+
return next(self.parameters()).dtype
|
| 178 |
+
|
| 179 |
+
@property
|
| 180 |
+
def device(self):
|
| 181 |
+
return next(self.parameters()).device
|
| 182 |
+
|
| 183 |
+
def _flatten_navit(self, pixel_values: torch.Tensor):
|
| 184 |
+
"""[N, 3, H, W] → (flat=[N*L, patch_dim], grid_thw=[N, 3], shape=(N, L)).
|
| 185 |
+
|
| 186 |
+
L = grid_t * grid_h * grid_w = 1 * (H/16) * (W/16)
|
| 187 |
+
patch_dim = C * temporal_patch_size * patch_size^2 = 3 * 2 * 16 * 16 = 1536
|
| 188 |
+
"""
|
| 189 |
+
pixel_values = pixel_values.to(dtype=self.dtype)
|
| 190 |
+
tps = self.config.temporal_patch_size
|
| 191 |
+
ps = self.config.patch_size
|
| 192 |
+
ms = self.config.spatial_merge_size
|
| 193 |
+
|
| 194 |
+
patches = pixel_values.unsqueeze(1)
|
| 195 |
+
# 对单帧图像 T=1, pad=1 → expand 一帧使 T 整除 temporal_patch_size，
|
| 196 |
+
# Conv3d 在复制帧上退化为等效 2D Conv（数学无损）
|
| 197 |
+
T = patches.shape[1]
|
| 198 |
+
pad = -T % tps
|
| 199 |
+
if pad:
|
| 200 |
+
repeats = patches[:, -1:].expand(-1, pad, -1, -1, -1)
|
| 201 |
+
patches = torch.cat((patches, repeats), dim=1)
|
| 202 |
+
|
| 203 |
+
batch_size, t, channel, H, W = patches.shape
|
| 204 |
+
grid_t = t // tps
|
| 205 |
+
grid_h = H // ps
|
| 206 |
+
grid_w = W // ps
|
| 207 |
+
|
| 208 |
+
patches = patches.view(
|
| 209 |
+
batch_size, grid_t, tps, channel,
|
| 210 |
+
grid_h // ms, ms, ps,
|
| 211 |
+
grid_w // ms, ms, ps,
|
| 212 |
+
)
|
| 213 |
+
patches = patches.permute(0, 1, 4, 7, 5, 8, 3, 2, 6, 9)
|
| 214 |
+
flatten_patches = patches.reshape(
|
| 215 |
+
batch_size,
|
| 216 |
+
grid_t * grid_h * grid_w,
|
| 217 |
+
channel * tps * ps * ps,
|
| 218 |
+
)
|
| 219 |
+
|
| 220 |
+
seq_len = grid_t * grid_h * grid_w
|
| 221 |
+
flat = flatten_patches.reshape(batch_size * seq_len, -1)
|
| 222 |
+
# on-device 构造小 tensor 再 expand，host→GPU 同步量 O(3) 而非 O(N*3)
|
| 223 |
+
grid_unit = torch.tensor(
|
| 224 |
+
[grid_t, grid_h, grid_w], dtype=torch.int32, device=pixel_values.device,
|
| 225 |
+
)
|
| 226 |
+
grid_thw = grid_unit.unsqueeze(0).expand(batch_size, -1).contiguous()
|
| 227 |
+
return flat, grid_thw, (batch_size, seq_len)
|
| 228 |
+
|
| 229 |
+
def forward(
|
| 230 |
+
self,
|
| 231 |
+
pixel_values: torch.Tensor,
|
| 232 |
+
output_hidden_states: Optional[bool] = None,
|
| 233 |
+
return_dict: Optional[bool] = None,
|
| 234 |
+
**kwargs,
|
| 235 |
+
) -> BaseModelOutputWithPooling:
|
| 236 |
+
flat, grid_thw, (N, S) = self._flatten_navit(pixel_values)
|
| 237 |
+
vision_out = self.vision(flat, grid_thw=grid_thw)
|
| 238 |
+
hidden = vision_out.last_hidden_state.view(N, S, -1)
|
| 239 |
+
|
| 240 |
+
# Block-merge → row-major reorder before handing to LlavaOnevision.
|
| 241 |
+
# Internally the ViT runs in Qwen NaViT block-merge layout (pretrained
|
| 242 |
+
# pos_embed + RoPE contract); downstream LlavaOV `pack_image_features`
|
| 243 |
+
# (multi-tile AnyRes path, view(num_patch_h, num_patch_w, h, w, -1))
|
| 244 |
+
# and `apply_pooling` (video path, view(B, h, w, -1) + bilinear) BOTH
|
| 245 |
+
# assume row-major. Without this reorder, the multi-tile/video spatial
|
| 246 |
+
# pool pulls together tokens that are NOT spatially adjacent — silent
|
| 247 |
+
# corruption that doesn't fire on S1 single-tile path (line 348-351 of
|
| 248 |
+
# modeling_llava_onevision.py just flattens [N,D] verbatim) but kills
|
| 249 |
+
# S2 / eval quality.
|
| 250 |
+
grid_h = int(grid_thw[0, 1].item())
|
| 251 |
+
grid_w = int(grid_thw[0, 2].item())
|
| 252 |
+
ms = getattr(self.config, "spatial_merge_size", 2)
|
| 253 |
+
layout_perm = block_merge_to_row_major_permutation(
|
| 254 |
+
grid_h, grid_w, ms=ms, device=hidden.device,
|
| 255 |
+
)
|
| 256 |
+
hidden = hidden[:, layout_perm, :]
|
| 257 |
+
|
| 258 |
+
return BaseModelOutputWithPooling(
|
| 259 |
+
last_hidden_state=hidden,
|
| 260 |
+
# LlavaOnevision 索引 hidden_states[vision_feature_layer=-1]；tuple 长度 1 足够
|
| 261 |
+
hidden_states=(hidden,),
|
| 262 |
+
pooler_output=None,
|
| 263 |
+
)
|
| 264 |
+
|
| 265 |
+
|
| 266 |
+
class LlavaQwen3_5ViTMultiModalProjector(nn.Module):
|
| 267 |
+
"""标准 LlavaOnevision projector（V6 final_layernorm fix (2026-05-16): pre_norm → Identity）。
|
| 268 |
+
|
| 269 |
+
V6 final_layernorm fix (2026-05-16) 修复后, encoder 末端已自带 final_layernorm（与 SigLIP2 post_layernorm 对称），
|
| 270 |
+
projector 不再需要补 LN — pre_norm 改为 nn.Identity，对齐 SigLIP2 plugin 的
|
| 271 |
+
LlavaOnevision stock projector 结构（裸 linear_1 → GELU → linear_2），
|
| 272 |
+
保证 SigLIP2 / Qwen3.5 / Qwen3-VL 三个 backbone 在 LlavaOV 设定下公平对比。
|
| 273 |
+
|
| 274 |
+
（历史：V6.0.0~V6.0.4 时期 encoder 无 final LN，projector pre_norm 是补丁；
|
| 275 |
+
现在补丁回到 encoder 内部，projector 回归 stock 形态。）
|
| 276 |
+
"""
|
| 277 |
+
|
| 278 |
+
def __init__(self, config: LlavaQwen3_5ViTConfig):
|
| 279 |
+
super().__init__()
|
| 280 |
+
num_feature_layers = (
|
| 281 |
+
1 if isinstance(config.vision_feature_layer, int) else len(config.vision_feature_layer)
|
| 282 |
+
)
|
| 283 |
+
vision_dim = config.vision_config.hidden_size * num_feature_layers
|
| 284 |
+
text_dim = config.text_config.hidden_size
|
| 285 |
+
bias = getattr(config, "multimodal_projector_bias", True)
|
| 286 |
+
|
| 287 |
+
# V6 final_layernorm fix (2026-05-16): pre_norm = Identity (encoder 已自带 final_layernorm).
|
| 288 |
+
self.pre_norm = nn.Identity()
|
| 289 |
+
self.linear_1 = nn.Linear(vision_dim, text_dim, bias=bias)
|
| 290 |
+
self.act = ACT2FN[config.projector_hidden_act]
|
| 291 |
+
self.linear_2 = nn.Linear(text_dim, text_dim, bias=bias)
|
| 292 |
+
|
| 293 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
| 294 |
+
return self.linear_2(self.act(self.linear_1(self.pre_norm(x))))
|
| 295 |
+
|
| 296 |
+
|
| 297 |
+
class LlavaQwen3_5ViTModel(LlavaOnevisionModel):
|
| 298 |
+
"""继承 LlavaOnevisionModel 但绕过其 __init__ 手动装配。
|
| 299 |
+
|
| 300 |
+
父类 __init__ 调 `AutoModel.from_config(config.vision_config)` 会对 Qwen3_5VisionConfig
|
| 301 |
+
抛 "Unrecognized configuration"（Qwen3.5 vision 没注册到 AutoModel）。手动装配避开
|
| 302 |
+
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）。
|
| 303 |
+
|
| 304 |
+
装配顺序与父类一致：vision_tower / projector / image_newline / language_model / post_init。
|
| 305 |
+
"""
|
| 306 |
+
|
| 307 |
+
config_class = LlavaQwen3_5ViTConfig
|
| 308 |
+
|
| 309 |
+
def __init__(self, config: LlavaQwen3_5ViTConfig):
|
| 310 |
+
# 跳过 LlavaOnevisionModel.__init__（AutoModel 不识别 Qwen3_5VisionConfig）
|
| 311 |
+
LlavaOnevisionPreTrainedModel.__init__(self, config)
|
| 312 |
+
self.vision_tower = Qwen3_5ViTAsSiglipAdapter(config.vision_config)
|
| 313 |
+
self.multi_modal_projector = LlavaQwen3_5ViTMultiModalProjector(config)
|
| 314 |
+
embed_std = 1 / math.sqrt(config.text_config.hidden_size)
|
| 315 |
+
self.image_newline = nn.Parameter(
|
| 316 |
+
torch.randn(config.text_config.hidden_size, dtype=self.dtype) * embed_std
|
| 317 |
+
)
|
| 318 |
+
self.vocab_size = config.text_config.vocab_size
|
| 319 |
+
self.language_model = AutoModel.from_config(config.text_config)
|
| 320 |
+
self.post_init()
|
| 321 |
+
|
| 322 |
+
|
| 323 |
+
class LlavaQwen3_5ViTForConditionalGeneration(LlavaOnevisionForConditionalGeneration):
|
| 324 |
+
"""继承 LlavaOnevisionForConditionalGeneration，只换 self.model。
|
| 325 |
+
|
| 326 |
+
同样跳过父类 __init__（避免重复构造 LlavaOnevisionModel，根因见 LlavaQwen3_5ViTModel）。
|
| 327 |
+
"""
|
| 328 |
+
|
| 329 |
+
config_class = LlavaQwen3_5ViTConfig
|
| 330 |
+
|
| 331 |
+
def __init__(self, config: LlavaQwen3_5ViTConfig):
|
| 332 |
+
LlavaOnevisionPreTrainedModel.__init__(self, config)
|
| 333 |
+
self.model = LlavaQwen3_5ViTModel(config)
|
| 334 |
+
self.lm_head = nn.Linear(
|
| 335 |
+
config.text_config.hidden_size, config.text_config.vocab_size, bias=False
|
| 336 |
+
)
|
| 337 |
+
self.post_init()
|
| 338 |
+
|
| 339 |
+
|
| 340 |
+
AutoConfig.register(LlavaQwen3_5ViTConfig.model_type, LlavaQwen3_5ViTConfig)
|
| 341 |
+
AutoModelForCausalLM.register(LlavaQwen3_5ViTConfig, LlavaQwen3_5ViTForConditionalGeneration)
|
| 342 |
+
|
| 343 |
+
|
| 344 |
+
__all__ = [
|
| 345 |
+
"LlavaQwen3_5ViTConfig",
|
| 346 |
+
"Qwen3_5ViTBackbone",
|
| 347 |
+
"Qwen3_5ViTAsSiglipAdapter",
|
| 348 |
+
"LlavaQwen3_5ViTMultiModalProjector",
|
| 349 |
+
"LlavaQwen3_5ViTModel",
|
| 350 |
+
"LlavaQwen3_5ViTForConditionalGeneration",
|
| 351 |
+
]
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/processor_config.json
ADDED
|
@@ -0,0 +1,203 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"image_processor": {
|
| 3 |
+
"do_convert_rgb": true,
|
| 4 |
+
"do_normalize": true,
|
| 5 |
+
"do_pad": true,
|
| 6 |
+
"do_rescale": true,
|
| 7 |
+
"do_resize": true,
|
| 8 |
+
"image_grid_pinpoints": [
|
| 9 |
+
[
|
| 10 |
+
384,
|
| 11 |
+
384
|
| 12 |
+
],
|
| 13 |
+
[
|
| 14 |
+
384,
|
| 15 |
+
768
|
| 16 |
+
],
|
| 17 |
+
[
|
| 18 |
+
384,
|
| 19 |
+
1152
|
| 20 |
+
],
|
| 21 |
+
[
|
| 22 |
+
384,
|
| 23 |
+
1536
|
| 24 |
+
],
|
| 25 |
+
[
|
| 26 |
+
384,
|
| 27 |
+
1920
|
| 28 |
+
],
|
| 29 |
+
[
|
| 30 |
+
384,
|
| 31 |
+
2304
|
| 32 |
+
],
|
| 33 |
+
[
|
| 34 |
+
768,
|
| 35 |
+
384
|
| 36 |
+
],
|
| 37 |
+
[
|
| 38 |
+
768,
|
| 39 |
+
768
|
| 40 |
+
],
|
| 41 |
+
[
|
| 42 |
+
768,
|
| 43 |
+
1152
|
| 44 |
+
],
|
| 45 |
+
[
|
| 46 |
+
768,
|
| 47 |
+
1536
|
| 48 |
+
],
|
| 49 |
+
[
|
| 50 |
+
768,
|
| 51 |
+
1920
|
| 52 |
+
],
|
| 53 |
+
[
|
| 54 |
+
768,
|
| 55 |
+
2304
|
| 56 |
+
],
|
| 57 |
+
[
|
| 58 |
+
1152,
|
| 59 |
+
384
|
| 60 |
+
],
|
| 61 |
+
[
|
| 62 |
+
1152,
|
| 63 |
+
768
|
| 64 |
+
],
|
| 65 |
+
[
|
| 66 |
+
1152,
|
| 67 |
+
1152
|
| 68 |
+
],
|
| 69 |
+
[
|
| 70 |
+
1152,
|
| 71 |
+
1536
|
| 72 |
+
],
|
| 73 |
+
[
|
| 74 |
+
1152,
|
| 75 |
+
1920
|
| 76 |
+
],
|
| 77 |
+
[
|
| 78 |
+
1152,
|
| 79 |
+
2304
|
| 80 |
+
],
|
| 81 |
+
[
|
| 82 |
+
1536,
|
| 83 |
+
384
|
| 84 |
+
],
|
| 85 |
+
[
|
| 86 |
+
1536,
|
| 87 |
+
768
|
| 88 |
+
],
|
| 89 |
+
[
|
| 90 |
+
1536,
|
| 91 |
+
1152
|
| 92 |
+
],
|
| 93 |
+
[
|
| 94 |
+
1536,
|
| 95 |
+
1536
|
| 96 |
+
],
|
| 97 |
+
[
|
| 98 |
+
1536,
|
| 99 |
+
1920
|
| 100 |
+
],
|
| 101 |
+
[
|
| 102 |
+
1536,
|
| 103 |
+
2304
|
| 104 |
+
],
|
| 105 |
+
[
|
| 106 |
+
1920,
|
| 107 |
+
384
|
| 108 |
+
],
|
| 109 |
+
[
|
| 110 |
+
1920,
|
| 111 |
+
768
|
| 112 |
+
],
|
| 113 |
+
[
|
| 114 |
+
1920,
|
| 115 |
+
1152
|
| 116 |
+
],
|
| 117 |
+
[
|
| 118 |
+
1920,
|
| 119 |
+
1536
|
| 120 |
+
],
|
| 121 |
+
[
|
| 122 |
+
1920,
|
| 123 |
+
1920
|
| 124 |
+
],
|
| 125 |
+
[
|
| 126 |
+
1920,
|
| 127 |
+
2304
|
| 128 |
+
],
|
| 129 |
+
[
|
| 130 |
+
2304,
|
| 131 |
+
384
|
| 132 |
+
],
|
| 133 |
+
[
|
| 134 |
+
2304,
|
| 135 |
+
768
|
| 136 |
+
],
|
| 137 |
+
[
|
| 138 |
+
2304,
|
| 139 |
+
1152
|
| 140 |
+
],
|
| 141 |
+
[
|
| 142 |
+
2304,
|
| 143 |
+
1536
|
| 144 |
+
],
|
| 145 |
+
[
|
| 146 |
+
2304,
|
| 147 |
+
1920
|
| 148 |
+
],
|
| 149 |
+
[
|
| 150 |
+
2304,
|
| 151 |
+
2304
|
| 152 |
+
]
|
| 153 |
+
],
|
| 154 |
+
"image_mean": [
|
| 155 |
+
0.5,
|
| 156 |
+
0.5,
|
| 157 |
+
0.5
|
| 158 |
+
],
|
| 159 |
+
"image_processor_type": "LlavaOnevisionImageProcessor",
|
| 160 |
+
"image_std": [
|
| 161 |
+
0.5,
|
| 162 |
+
0.5,
|
| 163 |
+
0.5
|
| 164 |
+
],
|
| 165 |
+
"resample": 3,
|
| 166 |
+
"rescale_factor": 0.00392156862745098,
|
| 167 |
+
"size": {
|
| 168 |
+
"height": 384,
|
| 169 |
+
"width": 384
|
| 170 |
+
}
|
| 171 |
+
},
|
| 172 |
+
"image_token": "<image>",
|
| 173 |
+
"num_image_tokens": 576,
|
| 174 |
+
"processor_class": "LlavaOnevisionProcessor",
|
| 175 |
+
"video_processor": {
|
| 176 |
+
"do_convert_rgb": true,
|
| 177 |
+
"do_normalize": true,
|
| 178 |
+
"do_rescale": true,
|
| 179 |
+
"do_resize": true,
|
| 180 |
+
"do_sample_frames": false,
|
| 181 |
+
"image_mean": [
|
| 182 |
+
0.5,
|
| 183 |
+
0.5,
|
| 184 |
+
0.5
|
| 185 |
+
],
|
| 186 |
+
"image_std": [
|
| 187 |
+
0.5,
|
| 188 |
+
0.5,
|
| 189 |
+
0.5
|
| 190 |
+
],
|
| 191 |
+
"resample": 3,
|
| 192 |
+
"rescale_factor": 0.00392156862745098,
|
| 193 |
+
"return_metadata": false,
|
| 194 |
+
"size": {
|
| 195 |
+
"height": 384,
|
| 196 |
+
"width": 384
|
| 197 |
+
},
|
| 198 |
+
"video_processor_type": "LlavaOnevisionVideoProcessor"
|
| 199 |
+
},
|
| 200 |
+
"video_token": "<video>",
|
| 201 |
+
"vision_aspect_ratio": "anyres_max_9",
|
| 202 |
+
"vision_feature_select_strategy": null
|
| 203 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9024318c850eaadf26be79389d21b07a7afd8f1af749b89f9109b06c0d12173c
|
| 3 |
+
size 11423018
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/tokenizer_config.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": null,
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|im_end|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<image>",
|
| 10 |
+
"<video>"
|
| 11 |
+
],
|
| 12 |
+
"is_local": true,
|
| 13 |
+
"model_max_length": 1010000,
|
| 14 |
+
"pad_token": "<|endoftext|>",
|
| 15 |
+
"processor_class": "LlavaOnevisionProcessor",
|
| 16 |
+
"split_special_tokens": false,
|
| 17 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 18 |
+
"unk_token": null
|
| 19 |
+
}
|
ckpts/S2/4b/qwen3_5_2b/v9_1_16f_10pct/v0-20260602-181529/checkpoint-505/zero_to_fp32.py
ADDED
|
@@ -0,0 +1,760 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python
|
| 2 |
+
|
| 3 |
+
# Copyright (c) Microsoft Corporation.
|
| 4 |
+
# SPDX-License-Identifier: Apache-2.0
|
| 5 |
+
|
| 6 |
+
# DeepSpeed Team
|
| 7 |
+
|
| 8 |
+
# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
|
| 9 |
+
# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
|
| 10 |
+
# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
|
| 11 |
+
# application.
|
| 12 |
+
#
|
| 13 |
+
# example:
|
| 14 |
+
# python zero_to_fp32.py . output_dir/
|
| 15 |
+
# or
|
| 16 |
+
# python zero_to_fp32.py . output_dir/ --safe_serialization
|
| 17 |
+
|
| 18 |
+
import argparse
|
| 19 |
+
import torch
|
| 20 |
+
import glob
|
| 21 |
+
import math
|
| 22 |
+
import os
|
| 23 |
+
import re
|
| 24 |
+
import gc
|
| 25 |
+
import json
|
| 26 |
+
import numpy as np
|
| 27 |
+
from tqdm import tqdm
|
| 28 |
+
from collections import OrderedDict
|
| 29 |
+
from dataclasses import dataclass
|
| 30 |
+
|
| 31 |
+
# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
|
| 32 |
+
# DeepSpeed data structures it has to be available in the current python environment.
|
| 33 |
+
from deepspeed.utils import logger
|
| 34 |
+
from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
|
| 35 |
+
FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
|
| 36 |
+
FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
@dataclass
|
| 40 |
+
class zero_model_state:
|
| 41 |
+
buffers: dict()
|
| 42 |
+
param_shapes: dict()
|
| 43 |
+
shared_params: list
|
| 44 |
+
ds_version: int
|
| 45 |
+
frozen_param_shapes: dict()
|
| 46 |
+
frozen_param_fragments: dict()
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
debug = 0
|
| 50 |
+
|
| 51 |
+
# load to cpu
|
| 52 |
+
device = torch.device('cpu')
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
def atoi(text):
|
| 56 |
+
return int(text) if text.isdigit() else text
|
| 57 |
+
|
| 58 |
+
|
| 59 |
+
def natural_keys(text):
|
| 60 |
+
'''
|
| 61 |
+
alist.sort(key=natural_keys) sorts in human order
|
| 62 |
+
http://nedbatchelder.com/blog/200712/human_sorting.html
|
| 63 |
+
(See Toothy's implementation in the comments)
|
| 64 |
+
'''
|
| 65 |
+
return [atoi(c) for c in re.split(r'(\d+)', text)]
|
| 66 |
+
|
| 67 |
+
|
| 68 |
+
def get_model_state_file(checkpoint_dir, zero_stage):
|
| 69 |
+
if not os.path.isdir(checkpoint_dir):
|
| 70 |
+
raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
|
| 71 |
+
|
| 72 |
+
# there should be only one file
|
| 73 |
+
if zero_stage <= 2:
|
| 74 |
+
file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
|
| 75 |
+
elif zero_stage == 3:
|
| 76 |
+
file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
|
| 77 |
+
|
| 78 |
+
if not os.path.exists(file):
|
| 79 |
+
raise FileNotFoundError(f"can't find model states file at '{file}'")
|
| 80 |
+
|
| 81 |
+
return file
|
| 82 |
+
|
| 83 |
+
|
| 84 |
+
def get_checkpoint_files(checkpoint_dir, glob_pattern):
|
| 85 |
+
# XXX: need to test that this simple glob rule works for multi-node setup too
|
| 86 |
+
ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
|
| 87 |
+
|
| 88 |
+
if len(ckpt_files) == 0:
|
| 89 |
+
raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
|
| 90 |
+
|
| 91 |
+
return ckpt_files
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
def get_optim_files(checkpoint_dir):
|
| 95 |
+
return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
|
| 96 |
+
|
| 97 |
+
|
| 98 |
+
def get_model_state_files(checkpoint_dir):
|
| 99 |
+
return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
def parse_model_states(files):
|
| 103 |
+
zero_model_states = []
|
| 104 |
+
for file in files:
|
| 105 |
+
state_dict = torch.load(file, map_location=device, weights_only=False)
|
| 106 |
+
|
| 107 |
+
if BUFFER_NAMES not in state_dict:
|
| 108 |
+
raise ValueError(f"{file} is not a model state checkpoint")
|
| 109 |
+
buffer_names = state_dict[BUFFER_NAMES]
|
| 110 |
+
if debug:
|
| 111 |
+
print("Found buffers:", buffer_names)
|
| 112 |
+
|
| 113 |
+
# recover just the buffers while restoring them to fp32 if they were saved in fp16
|
| 114 |
+
buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
|
| 115 |
+
param_shapes = state_dict[PARAM_SHAPES]
|
| 116 |
+
|
| 117 |
+
# collect parameters that are included in param_shapes
|
| 118 |
+
param_names = []
|
| 119 |
+
for s in param_shapes:
|
| 120 |
+
for name in s.keys():
|
| 121 |
+
param_names.append(name)
|
| 122 |
+
|
| 123 |
+
# update with frozen parameters
|
| 124 |
+
frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
|
| 125 |
+
if frozen_param_shapes is not None:
|
| 126 |
+
if debug:
|
| 127 |
+
print(f"Found frozen_param_shapes: {frozen_param_shapes}")
|
| 128 |
+
param_names += list(frozen_param_shapes.keys())
|
| 129 |
+
|
| 130 |
+
# handle shared params
|
| 131 |
+
shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
|
| 132 |
+
|
| 133 |
+
ds_version = state_dict.get(DS_VERSION, None)
|
| 134 |
+
|
| 135 |
+
frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
|
| 136 |
+
|
| 137 |
+
z_model_state = zero_model_state(buffers=buffers,
|
| 138 |
+
param_shapes=param_shapes,
|
| 139 |
+
shared_params=shared_params,
|
| 140 |
+
ds_version=ds_version,
|
| 141 |
+
frozen_param_shapes=frozen_param_shapes,
|
| 142 |
+
frozen_param_fragments=frozen_param_fragments)
|
| 143 |
+
zero_model_states.append(z_model_state)
|
| 144 |
+
|
| 145 |
+
return zero_model_states
|
| 146 |
+
|
| 147 |
+
|
| 148 |
+
def parse_optim_states(files, ds_checkpoint_dir):
|
| 149 |
+
total_files = len(files)
|
| 150 |
+
state_dicts = []
|
| 151 |
+
for f in tqdm(files, desc='Loading checkpoint shards'):
|
| 152 |
+
state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
|
| 153 |
+
# immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
|
| 154 |
+
# and also handle the case where it was already removed by another helper script
|
| 155 |
+
state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
|
| 156 |
+
state_dicts.append(state_dict)
|
| 157 |
+
|
| 158 |
+
if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
|
| 159 |
+
raise ValueError(f"{files[0]} is not a zero checkpoint")
|
| 160 |
+
zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
|
| 161 |
+
world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
|
| 162 |
+
|
| 163 |
+
# For ZeRO-2 each param group can have different partition_count as data parallelism for expert
|
| 164 |
+
# parameters can be different from data parallelism for non-expert parameters. So we can just
|
| 165 |
+
# use the max of the partition_count to get the dp world_size.
|
| 166 |
+
|
| 167 |
+
if type(world_size) is list:
|
| 168 |
+
world_size = max(world_size)
|
| 169 |
+
|
| 170 |
+
if world_size != total_files:
|
| 171 |
+
raise ValueError(
|
| 172 |
+
f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
|
| 173 |
+
"Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
|
| 174 |
+
)
|
| 175 |
+
|
| 176 |
+
# the groups are named differently in each stage
|
| 177 |
+
if zero_stage <= 2:
|
| 178 |
+
fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
|
| 179 |
+
elif zero_stage == 3:
|
| 180 |
+
fp32_groups_key = FP32_FLAT_GROUPS
|
| 181 |
+
else:
|
| 182 |
+
raise ValueError(f"unknown zero stage {zero_stage}")
|
| 183 |
+
|
| 184 |
+
fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
|
| 185 |
+
return zero_stage, world_size, fp32_flat_groups
|
| 186 |
+
|
| 187 |
+
|
| 188 |
+
def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
|
| 189 |
+
"""
|
| 190 |
+
Returns fp32 state_dict reconstructed from ds checkpoint
|
| 191 |
+
|
| 192 |
+
Args:
|
| 193 |
+
- ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
|
| 194 |
+
|
| 195 |
+
"""
|
| 196 |
+
print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
|
| 197 |
+
|
| 198 |
+
optim_files = get_optim_files(ds_checkpoint_dir)
|
| 199 |
+
zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
|
| 200 |
+
print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
|
| 201 |
+
|
| 202 |
+
model_files = get_model_state_files(ds_checkpoint_dir)
|
| 203 |
+
|
| 204 |
+
zero_model_states = parse_model_states(model_files)
|
| 205 |
+
print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
|
| 206 |
+
|
| 207 |
+
if zero_stage <= 2:
|
| 208 |
+
return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
| 209 |
+
exclude_frozen_parameters)
|
| 210 |
+
elif zero_stage == 3:
|
| 211 |
+
return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
| 212 |
+
exclude_frozen_parameters)
|
| 213 |
+
|
| 214 |
+
|
| 215 |
+
def _zero2_merge_frozen_params(state_dict, zero_model_states):
|
| 216 |
+
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
| 217 |
+
return
|
| 218 |
+
|
| 219 |
+
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
| 220 |
+
frozen_param_fragments = zero_model_states[0].frozen_param_fragments
|
| 221 |
+
|
| 222 |
+
if debug:
|
| 223 |
+
num_elem = sum(s.numel() for s in frozen_param_shapes.values())
|
| 224 |
+
print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
| 225 |
+
|
| 226 |
+
wanted_params = len(frozen_param_shapes)
|
| 227 |
+
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
| 228 |
+
avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
|
| 229 |
+
print(f'Frozen params: Have {avail_numel} numels to process.')
|
| 230 |
+
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
| 231 |
+
|
| 232 |
+
total_params = 0
|
| 233 |
+
total_numel = 0
|
| 234 |
+
for name, shape in frozen_param_shapes.items():
|
| 235 |
+
total_params += 1
|
| 236 |
+
unpartitioned_numel = shape.numel()
|
| 237 |
+
total_numel += unpartitioned_numel
|
| 238 |
+
|
| 239 |
+
state_dict[name] = frozen_param_fragments[name]
|
| 240 |
+
|
| 241 |
+
if debug:
|
| 242 |
+
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
| 243 |
+
|
| 244 |
+
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
| 245 |
+
|
| 246 |
+
|
| 247 |
+
def _has_callable(obj, fn):
|
| 248 |
+
attr = getattr(obj, fn, None)
|
| 249 |
+
return callable(attr)
|
| 250 |
+
|
| 251 |
+
|
| 252 |
+
def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
| 253 |
+
param_shapes = zero_model_states[0].param_shapes
|
| 254 |
+
|
| 255 |
+
# Reconstruction protocol:
|
| 256 |
+
#
|
| 257 |
+
# XXX: document this
|
| 258 |
+
|
| 259 |
+
if debug:
|
| 260 |
+
for i in range(world_size):
|
| 261 |
+
for j in range(len(fp32_flat_groups[0])):
|
| 262 |
+
print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
|
| 263 |
+
|
| 264 |
+
# XXX: memory usage doubles here (zero2)
|
| 265 |
+
num_param_groups = len(fp32_flat_groups[0])
|
| 266 |
+
merged_single_partition_of_fp32_groups = []
|
| 267 |
+
for i in range(num_param_groups):
|
| 268 |
+
merged_partitions = [sd[i] for sd in fp32_flat_groups]
|
| 269 |
+
full_single_fp32_vector = torch.cat(merged_partitions, 0)
|
| 270 |
+
merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
|
| 271 |
+
avail_numel = sum(
|
| 272 |
+
[full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
|
| 273 |
+
|
| 274 |
+
if debug:
|
| 275 |
+
wanted_params = sum([len(shapes) for shapes in param_shapes])
|
| 276 |
+
wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
|
| 277 |
+
# not asserting if there is a mismatch due to possible padding
|
| 278 |
+
print(f"Have {avail_numel} numels to process.")
|
| 279 |
+
print(f"Need {wanted_numel} numels in {wanted_params} params.")
|
| 280 |
+
|
| 281 |
+
# params
|
| 282 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
| 283 |
+
# out-of-core computing solution
|
| 284 |
+
total_numel = 0
|
| 285 |
+
total_params = 0
|
| 286 |
+
for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
|
| 287 |
+
offset = 0
|
| 288 |
+
avail_numel = full_single_fp32_vector.numel()
|
| 289 |
+
for name, shape in shapes.items():
|
| 290 |
+
|
| 291 |
+
unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
|
| 292 |
+
total_numel += unpartitioned_numel
|
| 293 |
+
total_params += 1
|
| 294 |
+
|
| 295 |
+
if debug:
|
| 296 |
+
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
| 297 |
+
state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
|
| 298 |
+
offset += unpartitioned_numel
|
| 299 |
+
|
| 300 |
+
# Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
|
| 301 |
+
# avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
|
| 302 |
+
# paddings performed in the code it's almost impossible to predict the exact numbers w/o the
|
| 303 |
+
# live optimizer object, so we are checking that the numbers are within the right range
|
| 304 |
+
align_to = 2 * world_size
|
| 305 |
+
|
| 306 |
+
def zero2_align(x):
|
| 307 |
+
return align_to * math.ceil(x / align_to)
|
| 308 |
+
|
| 309 |
+
if debug:
|
| 310 |
+
print(f"original offset={offset}, avail_numel={avail_numel}")
|
| 311 |
+
|
| 312 |
+
offset = zero2_align(offset)
|
| 313 |
+
avail_numel = zero2_align(avail_numel)
|
| 314 |
+
|
| 315 |
+
if debug:
|
| 316 |
+
print(f"aligned offset={offset}, avail_numel={avail_numel}")
|
| 317 |
+
|
| 318 |
+
# Sanity check
|
| 319 |
+
if offset != avail_numel:
|
| 320 |
+
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
| 321 |
+
|
| 322 |
+
print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
|
| 323 |
+
|
| 324 |
+
|
| 325 |
+
def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
| 326 |
+
exclude_frozen_parameters):
|
| 327 |
+
state_dict = OrderedDict()
|
| 328 |
+
|
| 329 |
+
# buffers
|
| 330 |
+
buffers = zero_model_states[0].buffers
|
| 331 |
+
state_dict.update(buffers)
|
| 332 |
+
if debug:
|
| 333 |
+
print(f"added {len(buffers)} buffers")
|
| 334 |
+
|
| 335 |
+
if not exclude_frozen_parameters:
|
| 336 |
+
_zero2_merge_frozen_params(state_dict, zero_model_states)
|
| 337 |
+
|
| 338 |
+
_zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
| 339 |
+
|
| 340 |
+
# recover shared parameters
|
| 341 |
+
for pair in zero_model_states[0].shared_params:
|
| 342 |
+
if pair[1] in state_dict:
|
| 343 |
+
state_dict[pair[0]] = state_dict[pair[1]]
|
| 344 |
+
|
| 345 |
+
return state_dict
|
| 346 |
+
|
| 347 |
+
|
| 348 |
+
def zero3_partitioned_param_info(unpartitioned_numel, world_size):
|
| 349 |
+
remainder = unpartitioned_numel % world_size
|
| 350 |
+
padding_numel = (world_size - remainder) if remainder else 0
|
| 351 |
+
partitioned_numel = math.ceil(unpartitioned_numel / world_size)
|
| 352 |
+
return partitioned_numel, padding_numel
|
| 353 |
+
|
| 354 |
+
|
| 355 |
+
def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
|
| 356 |
+
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
| 357 |
+
return
|
| 358 |
+
|
| 359 |
+
if debug:
|
| 360 |
+
for i in range(world_size):
|
| 361 |
+
num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
|
| 362 |
+
print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
| 363 |
+
|
| 364 |
+
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
| 365 |
+
wanted_params = len(frozen_param_shapes)
|
| 366 |
+
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
| 367 |
+
avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
|
| 368 |
+
print(f'Frozen params: Have {avail_numel} numels to process.')
|
| 369 |
+
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
| 370 |
+
|
| 371 |
+
total_params = 0
|
| 372 |
+
total_numel = 0
|
| 373 |
+
for name, shape in zero_model_states[0].frozen_param_shapes.items():
|
| 374 |
+
total_params += 1
|
| 375 |
+
unpartitioned_numel = shape.numel()
|
| 376 |
+
total_numel += unpartitioned_numel
|
| 377 |
+
|
| 378 |
+
param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
|
| 379 |
+
state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
|
| 380 |
+
|
| 381 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
| 382 |
+
|
| 383 |
+
if debug:
|
| 384 |
+
print(
|
| 385 |
+
f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
| 386 |
+
)
|
| 387 |
+
|
| 388 |
+
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
| 389 |
+
|
| 390 |
+
|
| 391 |
+
class GatheredTensor:
|
| 392 |
+
"""
|
| 393 |
+
A pseudo tensor that collects partitioned weights.
|
| 394 |
+
It is more memory efficient when there are multiple groups.
|
| 395 |
+
"""
|
| 396 |
+
|
| 397 |
+
def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
|
| 398 |
+
self.flat_groups = flat_groups
|
| 399 |
+
self.flat_groups_offset = flat_groups_offset
|
| 400 |
+
self.offset = offset
|
| 401 |
+
self.partitioned_numel = partitioned_numel
|
| 402 |
+
self.shape = shape
|
| 403 |
+
self.dtype = self.flat_groups[0][0].dtype
|
| 404 |
+
|
| 405 |
+
def contiguous(self):
|
| 406 |
+
"""
|
| 407 |
+
Merge partitioned weights from flat_groups into a single tensor.
|
| 408 |
+
"""
|
| 409 |
+
end_idx = self.offset + self.partitioned_numel
|
| 410 |
+
world_size = len(self.flat_groups)
|
| 411 |
+
pad_flat_param_chunks = []
|
| 412 |
+
|
| 413 |
+
for rank_i in range(world_size):
|
| 414 |
+
# for each rank, we need to collect weights from related group/groups
|
| 415 |
+
flat_groups_at_rank_i = self.flat_groups[rank_i]
|
| 416 |
+
start_group_id = None
|
| 417 |
+
end_group_id = None
|
| 418 |
+
for group_id in range(len(self.flat_groups_offset)):
|
| 419 |
+
if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
|
| 420 |
+
start_group_id = group_id
|
| 421 |
+
if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
|
| 422 |
+
end_group_id = group_id
|
| 423 |
+
break
|
| 424 |
+
# collect weights from related group/groups
|
| 425 |
+
for group_id in range(start_group_id, end_group_id + 1):
|
| 426 |
+
flat_tensor = flat_groups_at_rank_i[group_id]
|
| 427 |
+
start_offset = self.offset - self.flat_groups_offset[group_id]
|
| 428 |
+
end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
|
| 429 |
+
pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
|
| 430 |
+
|
| 431 |
+
# collect weights from all ranks
|
| 432 |
+
pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
|
| 433 |
+
param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
|
| 434 |
+
return param
|
| 435 |
+
|
| 436 |
+
|
| 437 |
+
def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
| 438 |
+
param_shapes = zero_model_states[0].param_shapes
|
| 439 |
+
avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
|
| 440 |
+
|
| 441 |
+
# Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
|
| 442 |
+
# param, re-consolidating each param, while dealing with padding if any
|
| 443 |
+
|
| 444 |
+
# merge list of dicts, preserving order
|
| 445 |
+
param_shapes = {k: v for d in param_shapes for k, v in d.items()}
|
| 446 |
+
|
| 447 |
+
if debug:
|
| 448 |
+
for i in range(world_size):
|
| 449 |
+
print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
|
| 450 |
+
|
| 451 |
+
wanted_params = len(param_shapes)
|
| 452 |
+
wanted_numel = sum(shape.numel() for shape in param_shapes.values())
|
| 453 |
+
# not asserting if there is a mismatch due to possible padding
|
| 454 |
+
avail_numel = fp32_flat_groups[0].numel() * world_size
|
| 455 |
+
print(f"Trainable params: Have {avail_numel} numels to process.")
|
| 456 |
+
print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
|
| 457 |
+
|
| 458 |
+
# params
|
| 459 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
| 460 |
+
# out-of-core computing solution
|
| 461 |
+
offset = 0
|
| 462 |
+
total_numel = 0
|
| 463 |
+
total_params = 0
|
| 464 |
+
flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
|
| 465 |
+
for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
|
| 466 |
+
unpartitioned_numel = shape.numel()
|
| 467 |
+
total_numel += unpartitioned_numel
|
| 468 |
+
total_params += 1
|
| 469 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
| 470 |
+
|
| 471 |
+
if debug:
|
| 472 |
+
print(
|
| 473 |
+
f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
| 474 |
+
)
|
| 475 |
+
|
| 476 |
+
# memory efficient tensor
|
| 477 |
+
tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
|
| 478 |
+
state_dict[name] = tensor
|
| 479 |
+
offset += partitioned_numel
|
| 480 |
+
|
| 481 |
+
offset *= world_size
|
| 482 |
+
|
| 483 |
+
# Sanity check
|
| 484 |
+
if offset != avail_numel:
|
| 485 |
+
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
| 486 |
+
|
| 487 |
+
print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
|
| 488 |
+
|
| 489 |
+
|
| 490 |
+
def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
| 491 |
+
exclude_frozen_parameters):
|
| 492 |
+
state_dict = OrderedDict()
|
| 493 |
+
|
| 494 |
+
# buffers
|
| 495 |
+
buffers = zero_model_states[0].buffers
|
| 496 |
+
state_dict.update(buffers)
|
| 497 |
+
if debug:
|
| 498 |
+
print(f"added {len(buffers)} buffers")
|
| 499 |
+
|
| 500 |
+
if not exclude_frozen_parameters:
|
| 501 |
+
_zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
|
| 502 |
+
|
| 503 |
+
_zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
| 504 |
+
|
| 505 |
+
# recover shared parameters
|
| 506 |
+
for pair in zero_model_states[0].shared_params:
|
| 507 |
+
if pair[1] in state_dict:
|
| 508 |
+
state_dict[pair[0]] = state_dict[pair[1]]
|
| 509 |
+
|
| 510 |
+
return state_dict
|
| 511 |
+
|
| 512 |
+
|
| 513 |
+
def to_torch_tensor(state_dict, return_empty_tensor=False):
|
| 514 |
+
"""
|
| 515 |
+
Convert state_dict of GatheredTensor to torch tensor
|
| 516 |
+
"""
|
| 517 |
+
torch_state_dict = {}
|
| 518 |
+
converted_tensors = {}
|
| 519 |
+
for name, tensor in state_dict.items():
|
| 520 |
+
tensor_id = id(tensor)
|
| 521 |
+
if tensor_id in converted_tensors: # shared tensors
|
| 522 |
+
shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
|
| 523 |
+
torch_state_dict[name] = shared_tensor
|
| 524 |
+
else:
|
| 525 |
+
converted_tensors[tensor_id] = name
|
| 526 |
+
if return_empty_tensor:
|
| 527 |
+
torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
|
| 528 |
+
else:
|
| 529 |
+
torch_state_dict[name] = tensor.contiguous()
|
| 530 |
+
return torch_state_dict
|
| 531 |
+
|
| 532 |
+
|
| 533 |
+
def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
|
| 534 |
+
tag=None,
|
| 535 |
+
exclude_frozen_parameters=False,
|
| 536 |
+
lazy_mode=False):
|
| 537 |
+
"""
|
| 538 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
|
| 539 |
+
``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
|
| 540 |
+
via a model hub.
|
| 541 |
+
|
| 542 |
+
Args:
|
| 543 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder
|
| 544 |
+
- ``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``
|
| 545 |
+
- ``exclude_frozen_parameters``: exclude frozen parameters
|
| 546 |
+
- ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
|
| 547 |
+
Convert the pesduo tensor to torch tensor by ``.contiguous()``
|
| 548 |
+
|
| 549 |
+
Returns:
|
| 550 |
+
- pytorch ``state_dict``
|
| 551 |
+
|
| 552 |
+
A typical usage might be ::
|
| 553 |
+
|
| 554 |
+
from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
|
| 555 |
+
# do the training and checkpoint saving
|
| 556 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
|
| 557 |
+
model = model.cpu() # move to cpu
|
| 558 |
+
model.load_state_dict(state_dict)
|
| 559 |
+
# submit to model hub or save the model to share with others
|
| 560 |
+
|
| 561 |
+
In this example the ``model`` will no longer be usable in the deepspeed context of the same
|
| 562 |
+
application. i.e. you will need to re-initialize the deepspeed engine, since
|
| 563 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
| 564 |
+
|
| 565 |
+
If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
|
| 566 |
+
|
| 567 |
+
Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
|
| 568 |
+
You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
|
| 569 |
+
the checkpoint. Or you can load state_dict in lazy mode ::
|
| 570 |
+
|
| 571 |
+
from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
|
| 572 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
|
| 573 |
+
for name, lazy_tensor in state_dict.item():
|
| 574 |
+
tensor = lazy_tensor.contiguous() # to cpu
|
| 575 |
+
print(name, tensor)
|
| 576 |
+
# del tensor to release memory if it no longer in use
|
| 577 |
+
"""
|
| 578 |
+
if tag is None:
|
| 579 |
+
latest_path = os.path.join(checkpoint_dir, 'latest')
|
| 580 |
+
if os.path.isfile(latest_path):
|
| 581 |
+
with open(latest_path, 'r') as fd:
|
| 582 |
+
tag = fd.read().strip()
|
| 583 |
+
else:
|
| 584 |
+
raise ValueError(f"Unable to find 'latest' file at {latest_path}")
|
| 585 |
+
|
| 586 |
+
ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
|
| 587 |
+
|
| 588 |
+
if not os.path.isdir(ds_checkpoint_dir):
|
| 589 |
+
raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
|
| 590 |
+
|
| 591 |
+
state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
|
| 592 |
+
if lazy_mode:
|
| 593 |
+
return state_dict
|
| 594 |
+
else:
|
| 595 |
+
return to_torch_tensor(state_dict)
|
| 596 |
+
|
| 597 |
+
|
| 598 |
+
def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
|
| 599 |
+
output_dir,
|
| 600 |
+
max_shard_size="5GB",
|
| 601 |
+
safe_serialization=False,
|
| 602 |
+
tag=None,
|
| 603 |
+
exclude_frozen_parameters=False):
|
| 604 |
+
"""
|
| 605 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
|
| 606 |
+
loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
|
| 607 |
+
|
| 608 |
+
Args:
|
| 609 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
| 610 |
+
- ``output_dir``: directory to the pytorch fp32 state_dict output files
|
| 611 |
+
- ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
|
| 612 |
+
- ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
|
| 613 |
+
- ``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``
|
| 614 |
+
- ``exclude_frozen_parameters``: exclude frozen parameters
|
| 615 |
+
"""
|
| 616 |
+
|
| 617 |
+
# Dependency pre-check
|
| 618 |
+
if safe_serialization:
|
| 619 |
+
try:
|
| 620 |
+
from safetensors.torch import save_file
|
| 621 |
+
except ImportError:
|
| 622 |
+
print('If you want to use `safe_serialization`, please `pip install safetensors`')
|
| 623 |
+
raise
|
| 624 |
+
if max_shard_size is not None:
|
| 625 |
+
try:
|
| 626 |
+
from huggingface_hub import split_torch_state_dict_into_shards
|
| 627 |
+
except ImportError:
|
| 628 |
+
print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
|
| 629 |
+
raise
|
| 630 |
+
|
| 631 |
+
# Convert zero checkpoint to state_dict
|
| 632 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
|
| 633 |
+
tag,
|
| 634 |
+
exclude_frozen_parameters,
|
| 635 |
+
lazy_mode=True)
|
| 636 |
+
|
| 637 |
+
# Shard the model if it is too big.
|
| 638 |
+
weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
|
| 639 |
+
if max_shard_size is not None:
|
| 640 |
+
filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
|
| 641 |
+
# an memory-efficient approach for sharding
|
| 642 |
+
empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
|
| 643 |
+
state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
|
| 644 |
+
filename_pattern=filename_pattern,
|
| 645 |
+
max_shard_size=max_shard_size)
|
| 646 |
+
else:
|
| 647 |
+
from collections import namedtuple
|
| 648 |
+
StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
|
| 649 |
+
state_dict_split = StateDictSplit(is_sharded=False,
|
| 650 |
+
filename_to_tensors={weights_name: list(state_dict.keys())})
|
| 651 |
+
|
| 652 |
+
# Save the model by shard
|
| 653 |
+
os.makedirs(output_dir, exist_ok=True)
|
| 654 |
+
filename_to_tensors = state_dict_split.filename_to_tensors.items()
|
| 655 |
+
for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
|
| 656 |
+
shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
|
| 657 |
+
shard_state_dict = to_torch_tensor(shard_state_dict)
|
| 658 |
+
output_path = os.path.join(output_dir, shard_file)
|
| 659 |
+
if safe_serialization:
|
| 660 |
+
save_file(shard_state_dict, output_path, metadata={"format": "pt"})
|
| 661 |
+
else:
|
| 662 |
+
torch.save(shard_state_dict, output_path)
|
| 663 |
+
# release the memory of current shard
|
| 664 |
+
for tensor_name in list(shard_state_dict.keys()):
|
| 665 |
+
del state_dict[tensor_name]
|
| 666 |
+
del shard_state_dict[tensor_name]
|
| 667 |
+
del shard_state_dict
|
| 668 |
+
gc.collect()
|
| 669 |
+
|
| 670 |
+
# Save index if sharded
|
| 671 |
+
if state_dict_split.is_sharded:
|
| 672 |
+
index = {
|
| 673 |
+
"metadata": state_dict_split.metadata,
|
| 674 |
+
"weight_map": state_dict_split.tensor_to_filename,
|
| 675 |
+
}
|
| 676 |
+
save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
|
| 677 |
+
save_index_file = os.path.join(output_dir, save_index_file)
|
| 678 |
+
with open(save_index_file, "w", encoding="utf-8") as f:
|
| 679 |
+
content = json.dumps(index, indent=2, sort_keys=True) + "\n"
|
| 680 |
+
f.write(content)
|
| 681 |
+
|
| 682 |
+
|
| 683 |
+
def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
|
| 684 |
+
"""
|
| 685 |
+
1. Put the provided model to cpu
|
| 686 |
+
2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
|
| 687 |
+
3. Load it into the provided model
|
| 688 |
+
|
| 689 |
+
Args:
|
| 690 |
+
- ``model``: the model object to update
|
| 691 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
| 692 |
+
- ``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``
|
| 693 |
+
|
| 694 |
+
Returns:
|
| 695 |
+
- ``model`: modified model
|
| 696 |
+
|
| 697 |
+
Make sure you have plenty of CPU memory available before you call this function. If you don't
|
| 698 |
+
have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
|
| 699 |
+
conveniently placed for you in the checkpoint folder.
|
| 700 |
+
|
| 701 |
+
A typical usage might be ::
|
| 702 |
+
|
| 703 |
+
from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
|
| 704 |
+
model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
|
| 705 |
+
# submit to model hub or save the model to share with others
|
| 706 |
+
|
| 707 |
+
Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
|
| 708 |
+
of the same application. i.e. you will need to re-initialize the deepspeed engine, since
|
| 709 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
| 710 |
+
|
| 711 |
+
"""
|
| 712 |
+
logger.info("Extracting fp32 weights")
|
| 713 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
| 714 |
+
|
| 715 |
+
logger.info("Overwriting model with fp32 weights")
|
| 716 |
+
model = model.cpu()
|
| 717 |
+
model.load_state_dict(state_dict, strict=False)
|
| 718 |
+
|
| 719 |
+
return model
|
| 720 |
+
|
| 721 |
+
|
| 722 |
+
if __name__ == "__main__":
|
| 723 |
+
parser = argparse.ArgumentParser()
|
| 724 |
+
parser.add_argument("checkpoint_dir",
|
| 725 |
+
type=str,
|
| 726 |
+
help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
|
| 727 |
+
parser.add_argument("output_dir",
|
| 728 |
+
type=str,
|
| 729 |
+
help="directory to the pytorch fp32 state_dict output files"
|
| 730 |
+
"(e.g. path/checkpoint-12-output/)")
|
| 731 |
+
parser.add_argument(
|
| 732 |
+
"--max_shard_size",
|
| 733 |
+
type=str,
|
| 734 |
+
default="5GB",
|
| 735 |
+
help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
|
| 736 |
+
"lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
|
| 737 |
+
"We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
|
| 738 |
+
"without CPU OOM issues.")
|
| 739 |
+
parser.add_argument(
|
| 740 |
+
"--safe_serialization",
|
| 741 |
+
default=False,
|
| 742 |
+
action='store_true',
|
| 743 |
+
help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
|
| 744 |
+
parser.add_argument("-t",
|
| 745 |
+
"--tag",
|
| 746 |
+
type=str,
|
| 747 |
+
default=None,
|
| 748 |
+
help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
|
| 749 |
+
parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
|
| 750 |
+
parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
|
| 751 |
+
args = parser.parse_args()
|
| 752 |
+
|
| 753 |
+
debug = args.debug
|
| 754 |
+
|
| 755 |
+
convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
|
| 756 |
+
args.output_dir,
|
| 757 |
+
max_shard_size=args.max_shard_size,
|
| 758 |
+
safe_serialization=args.safe_serialization,
|
| 759 |
+
tag=args.tag,
|
| 760 |
+
exclude_frozen_parameters=args.exclude_frozen_parameters)
|