Update tiny models for HyperCLOVAXVisionV2ForConditionalGeneration

chat_template.jinja

@@ -0,0 +1,164 @@
+{%- set ns_img = namespace(count=0) %}
+{%- set ns_vid = namespace(count=0) %}
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {%- if messages[0].content is string %}
+            {%- if messages[0].content %}
+                {{- messages[0].content + '\n\n' }}
+            {%- endif %}
+        {%- elif messages[0].content is sequence %}
+            {%- for content_part in messages[0].content %}
+                {%- if content_part.type == 'text' %}
+                    {{- content_part.text + '\n\n' }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+    {%- endif %}
+    {{- '# Tools\n\n' }}
+    {{- 'You may call one or more functions to assist with the user query.\n\n' }}
+    {{- 'You are provided with function signatures within <tools></tools> XML tags:\n' }}
+    {{- '<tools>\n' }}
+    {%- for tool in tools %}
+        {{- tool | tojson(ensure_ascii=False) }}
+    {%- endfor %}
+    {{- '\n</tools>\n\n' }}
+    {{- 'For each function call, output the function name and arguments within the following XML format:\n' }}
+    {{- '<tool_call>{function-name}\n' }}
+    {{- '<arg_key>{arg-key-1}</arg_key>\n' }}
+    {{- '<arg_value>{arg-value-1}</arg_value>\n' }}
+    {{- '<arg_key>{arg-key-2}</arg_key>\n' }}
+    {{- '<arg_value>{arg-value-2}</arg_value>\n' }}
+    {{- '...\n' }}
+    {{- '</tool_call><|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' }}
+        {%- if messages[0].content is string %}
+            {{- messages[0].content }}
+        {%- elif messages[0].content is sequence %}
+            {%- for content_part in messages[0].content %}
+                {%- if content_part.type == 'text' %}
+                    {{- content_part.text }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+        {%- if messages[0].content %}
+            {{- '\n' }}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(last_user_index=-1) %}
+{%- for m in messages %}
+    {%- if m.role == 'user' %}
+        {%- set ns.last_user_index = loop.index0 %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- set content = message.content %}
+    {%- if (message.role == 'system' and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' }}
+        {%- if content is string %}
+            {{- content }}
+        {%- elif content is sequence %}
+            {%- for content_part in content %}
+                {%- if content_part.type == 'text' %}
+                    {{- content_part.text }}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>' + '\n' }}
+    {%- elif message.role == 'user' %}
+        {{- '<|im_start|>user\n' }}
+        {%- if message['content'] is string %}
+            {{- message['content'] + '<|im_end|>\n' }}
+        {%- elif message['content'] is sequence %}
+            {%- for content in message['content'] %}
+                {%- if not loop.first %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if content['type'] == 'image' or 'image' in content or 'image_url' in content %}
+                    {%- set image_id = 'image_%02d' % ns_img.count %}
+                    {%- set ns_img.count = ns_img.count + 1 %}
+                    {%- set mime_type = content.get('mime_type', 'image/jpeg') %}
+                    {{- '<|mime_start|>{"id": "' + image_id + '", "type": "' + mime_type + '", "filename": "' + content.get('filename', "a.jpg") + '"}<|mime_end|>\n' }}
+                    {{- '<|image_start|><|IMAGE_PAD|><|image_end|>' }}
+                {%- elif content['type'] == 'video' or 'video' in content or 'video_url' in content %}
+                    {%- set video_id = 'video_%02d' % ns_vid.count %}
+                    {%- set ns_vid.count = ns_vid.count + 1 %}
+                    {%- set mime_type = content.get('mime_type', 'video/mp4') %}
+                    {{- '<|mime_start|>{"id": "' + video_id + '", "type": "' + mime_type + '", "filename": "' + content.get('filename', "a.mp4") + '"}<|mime_end|>\n' }}
+                    {{- '<|video_aux_start|>다음 중 video_duration은 비디오 길이 정보입니다. 참고하여 답변하세요. {"video_duration": ' + (content.get('video_duration') | tojson if content.get('video_duration') else '<|video_duration|>') + '}<|video_aux_end|>\n'}}
+                    {{- '<|video_start|><|VIDEO_PAD|><|video_end|>'}}
+                {%- elif content['type'] == 'text' %}
+                    {{- content['text'] }}
+                {%- endif %}
+            {%- endfor %}
+            {{- '<|im_end|>\n'}}
+        {%- endif %}
+    {%- elif message.role == 'assistant' %}
+        {{- '<|im_start|>assistant\n' }}
+        {%- if message.content is string %}
+            {%- set reasoning_content = '' %}
+            {%- if message.reasoning_content is string %}
+                {%- set reasoning_content = message.reasoning_content %}
+            {%- else %}
+                {%- if '</think>' in content %}
+                    {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                    {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+                {%- endif %}
+            {%- endif %}
+            {%- if loop.index0 > ns.last_user_index %}
+                {%- if (loop.index0 > 0 and messages[loop.index0 - 1].role == 'user') or reasoning_content %}
+                    {{- '<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+                {%- else %}
+                    {{- content }}
+                {%- endif %}
+            {%- else %}
+                {{- content }}
+            {%- endif %}
+        {%- elif message.content is sequence %}
+            {%- for content_part in message.content %}
+                {%- if content_part.type == 'text' %}
+                    {{- content_part.text }}
+                    {{- '\n' if not loop.last else '' -}}
+                {%- endif %}
+            {%- endfor %}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if not loop.first or content %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>' + tool_call.name + '\n' }}
+                {%- set _args = tool_call.arguments %}
+                {%- for k, v in _args.items() %}
+                    {{- '<arg_key>' + k + '</arg_key>\n' }}
+                    {{- '<arg_value>' + (v | tojson(ensure_ascii=False) if v is not string else v) + '</arg_value>\n' }}
+                {%- endfor %}
+                {{- '</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == 'tool' %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != 'tool') %}
+            {{- '<|im_start|>tool' }}
+        {%- endif %}
+        {{- '\n<tool_response>' + message.get('name', '') + '\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != 'tool') %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n<think>\n' }}
+    {%- if skip_reasoning is defined and skip_reasoning is true %}
+        {{- '\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

config.json

@@ -0,0 +1,69 @@
+{
+  "architectures": [
+    "HyperCLOVAXVisionV2ForConditionalGeneration"
+  ],
+  "dtype": "float32",
+  "image_token_id": 3,
+  "model_type": "hyperclovax_vision_v2",
+  "pad_token_id": 0,
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attention_multiplier": 0.25,
+    "bos_token_id": 0,
+    "embedding_multiplier": 1.0,
+    "eos_token_id": 128001,
+    "head_dim": 16,
+    "hidden_act": "gelu",
+    "hidden_size": 32,
+    "initializer_range": 0.02,
+    "intermediate_size": 32,
+    "logits_scaling": 1.0,
+    "max_position_embeddings": 512,
+    "mlp_bias": false,
+    "model_type": "hyperclovax",
+    "num_attention_heads": 2,
+    "num_hidden_layers": 2,
+    "num_key_value_heads": 2,
+    "pad_token_id": 0,
+    "residual_multiplier": 1.0,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "rope_theta": 10000.0,
+      "rope_type": "default"
+    },
+    "tie_word_embeddings": false,
+    "use_cache": true,
+    "use_post_norm": true,
+    "vocab_size": 128256
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.18.0.dev0",
+  "video_token_id": 4,
+  "vision_config": {
+    "depth": 2,
+    "fullatt_block_indexes": [
+      7,
+      15,
+      23,
+      31
+    ],
+    "hidden_act": "gelu",
+    "hidden_size": 32,
+    "in_channels": 3,
+    "initializer_range": 0.02,
+    "intermediate_size": 32,
+    "model_type": "qwen2_5_vl_vision",
+    "num_heads": 4,
+    "out_hidden_size": 32,
+    "patch_size": 14,
+    "rope_parameters": {
+      "rope_theta": 10000.0,
+      "rope_type": "axial"
+    },
+    "spatial_merge_size": 1,
+    "temporal_patch_size": 2,
+    "tokens_per_second": 1,
+    "window_size": 112
+  }
+}

generation_config.json

@@ -0,0 +1,10 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 128001,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 0,
+  "transformers_version": "5.18.0.dev0",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1adfaf1e9e589f2bce5b1ccc27e7dd6352de4ef07b898735ad1e0545ff1a335
+size 33121552

preprocessor_config.json

@@ -0,0 +1,76 @@
+{
+  "discrete_image_end_token": "<|discrete_image_end|>",
+  "discrete_image_ratios": [
+    [
+      1,
+      1
+    ],
+    [
+      1,
+      2
+    ],
+    [
+      2,
+      1
+    ],
+    [
+      3,
+      4
+    ],
+    [
+      4,
+      3
+    ],
+    [
+      9,
+      16
+    ],
+    [
+      16,
+      9
+    ],
+    [
+      1,
+      3
+    ],
+    [
+      3,
+      1
+    ]
+  ],
+  "discrete_image_start_token": "<|discrete_image_start|>",
+  "discrete_image_token": "<|DISCRETE_IMAGE_PAD|>",
+  "discrete_token_size": 27,
+  "do_center_crop": false,
+  "do_convert_rgb": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_end_token": "<|image_end|>",
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_processor_type": "Qwen2VLImageProcessor",
+  "image_start_token": "<|image_start|>",
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "image_token": "<|IMAGE_PAD|>",
+  "merge_size": 2,
+  "patch_size": 14,
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "resize_mode": "smart_resize",
+  "size": {
+    "longest_edge": 2073600,
+    "shortest_edge": 3136
+  },
+  "temporal_patch_size": 2,
+  "use_discrete_token": false,
+  "vision_eof_token": "<|vision_eof|>",
+  "vision_eol_token": "<|vision_eol|>"
+}

tokenizer_config.json

@@ -0,0 +1,23 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "fix_mistral_regex": true,
+  "image_token": "<|IMAGE_PAD|>",
+  "is_local": true,
+  "local_files_only": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "model_specific_special_tokens": {
+    "image_token": "<|IMAGE_PAD|>",
+    "video_token": "<|VIDEO_PAD|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "processor_class": "Exaone4_5_Processor",
+  "sep_token": "<|endoftext|>",
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "<|endoftext|>",
+  "video_token": "<|VIDEO_PAD|>"
+}

video_preprocessor_config.json

@@ -0,0 +1,39 @@
+{
+  "do_convert_rgb": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "do_sample_frames": false,
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "max_frames": 120,
+  "max_video_tokens": 128000,
+  "merge_size": 2,
+  "min_frames": 4,
+  "patch_size": 14,
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "return_metadata": false,
+  "size": {
+    "longest_edge": 285768,
+    "shortest_edge": 3136
+  },
+  "temporal_patch_size": 2,
+  "use_audio_in_video": false,
+  "use_discrete_token": false,
+  "video_audio_token": "<|VIDEO_AUDIO_PAD|>",
+  "video_end_token": "<|video_end|>",
+  "video_processor_type": "Qwen2VLVideoProcessor",
+  "video_start_token": "<|video_start|>",
+  "video_token": "<|VIDEO_PAD|>",
+  "vision_eof_token": "<|vision_eof|>",
+  "vision_eol_token": "<|vision_eol|>"
+}