Update tiny models for DeepseekVLHybridForConditionalGeneration

chat_template.jinja

@@ -0,0 +1,5 @@
+{% set seps = ['
+
+', '<｜end▁of▁sentence｜>'] %}{% set i = 0 %}You are a helpful language and vision assistant. You are able to understand the visual content that the user provides, and assist the user with a variety of tasks using natural language.
+
+{% for message in messages %}{% if message['role']|lower == 'user' %}User: {% elif message['role']|lower == 'assistant' %}Assistant:{% if not (loop.last and not add_generation_prompt and message['content'][0]['type']=='text' and message['content'][0]['text']=='') %} {% endif %}{% else %}{{ message['role'].capitalize() }}: {% endif %}{% for content in message['content'] %}{% if content['type'] == 'image' %}<image_placeholder>{% elif content['type'] == 'text' %}{% set text = content['text'] %}{% if loop.first %}{% set text = text.lstrip() %}{% endif %}{% if loop.last %}{% set text = text.rstrip() %}{% endif %}{% if not loop.first and message['content'][loop.index0-1]['type'] == 'text' %}{{ ' ' + text }}{% else %}{{ text }}{% endif %}{% endif %}{% endfor %}{% if not loop.last or add_generation_prompt %}{% if message['role']|lower == 'user' %}{{ seps[0] }}{% else %}{{ seps[1] }}{% endif %}{% endif %}{% endfor %}{% if add_generation_prompt %}Assistant:{% endif %}

config.json

@@ -0,0 +1,77 @@
+{
+  "architectures": [
+    "DeepseekVLHybridForConditionalGeneration"
+  ],
+  "dtype": "float32",
+  "high_res_vision_config": {
+    "attention_dropout": 0.0,
+    "global_attn_indexes": [
+      0
+    ],
+    "hidden_act": "gelu",
+    "hidden_size": 16,
+    "image_size": 128,
+    "initializer_range": 1e-10,
+    "intermediate_size": 37,
+    "layer_norm_eps": 1e-06,
+    "mlp_dim": 24,
+    "mlp_ratio": 4.0,
+    "model_type": "sam_vision_model",
+    "num_attention_heads": 4,
+    "num_channels": 3,
+    "num_hidden_layers": 2,
+    "num_pos_feats": 128,
+    "output_channels": 4,
+    "patch_size": 32,
+    "qkv_bias": true,
+    "scale": 8,
+    "use_abs_pos": true,
+    "use_rel_pos": true,
+    "window_size": 14
+  },
+  "image_token_id": 98,
+  "model_type": "deepseek_vl_hybrid",
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "bos_token_id": 100000,
+    "eos_token_id": 100001,
+    "head_dim": 4,
+    "hidden_act": "silu",
+    "hidden_size": 16,
+    "initializer_range": 0.02,
+    "intermediate_size": 37,
+    "max_position_embeddings": 512,
+    "mlp_bias": false,
+    "model_type": "llama",
+    "num_attention_heads": 4,
+    "num_hidden_layers": 2,
+    "num_key_value_heads": 4,
+    "pad_token_id": 100001,
+    "pretraining_tp": 1,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "rope_theta": 10000.0,
+      "rope_type": "default"
+    },
+    "tie_word_embeddings": false,
+    "use_cache": true,
+    "vocab_size": 100016
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.16.0.dev0",
+  "vision_config": {
+    "attention_dropout": 0.0,
+    "hidden_act": "gelu",
+    "hidden_size": 16,
+    "image_size": 32,
+    "intermediate_size": 37,
+    "layer_norm_eps": 1e-06,
+    "model_type": "siglip_vision_model",
+    "num_attention_heads": 4,
+    "num_channels": 3,
+    "num_hidden_layers": 1,
+    "patch_size": 8,
+    "vision_use_head": false
+  }
+}

generation_config.json

@@ -0,0 +1,10 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 100001,
+  "transformers_version": "5.16.0.dev0",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80ff5313ab21dc4da411adc33053ac23e962745dc235a686f42607049c49b092
+size 6682120

preprocessor_config.json

@@ -0,0 +1,53 @@
+{
+  "background_color": [
+    127,
+    127,
+    127
+  ],
+  "crop_size": {
+    "height": 32,
+    "width": 32
+  },
+  "do_normalize": true,
+  "do_pad": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "high_res_background_color": [
+    127,
+    127,
+    127
+  ],
+  "high_res_image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "high_res_image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "high_res_resample": 3,
+  "high_res_size": {
+    "height": 1024,
+    "width": 1024
+  },
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_processor_type": "DeepseekVLHybridImageProcessor",
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "min_size": 14,
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "height": 32,
+    "width": 32
+  }
+}

tokenizer_config.json

@@ -0,0 +1,20 @@
+{
+  "add_prefix_space": null,
+  "backend": "tokenizers",
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "image_token": "<image_placeholder>",
+  "is_local": true,
+  "legacy": true,
+  "local_files_only": false,
+  "model_max_length": 16384,
+  "model_specific_special_tokens": {
+    "image_token": "<image_placeholder>"
+  },
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "processor_class": "DeepseekVLHybridProcessor",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "use_default_system_prompt": false
+}