Update tiny models for Ernie4_5_VLMoeForConditionalGeneration

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

chat_template.jinja

@@ -0,0 +1,81 @@
+
+{%- set image_count = namespace(value=0) -%}
+{%- set video_count = namespace(value=0) -%}
+{{- '<|begin_of_sentence|>' }}
+{%- for message in messages -%}
+    {%- if message.role in ['system', 'user'] -%}
+        {%- if message.role == 'user' -%}
+            {{- 'User: ' -}}
+        {%- endif -%}
+        {%- if message.content is string -%}
+            {{- message.content -}}
+        {%- else -%}
+            {%- for content_item in message.content -%}
+                {%- if content_item.type == 'text' -%}
+                    {{- content_item.text -}}
+                {%- elif content_item.type in ['image_url', 'image'] -%}
+                    {%- set image_count.value = image_count.value + 1 -%}
+                    Picture {{ image_count.value }}:<|IMAGE_START|><|IMAGE_PLACEHOLDER|><|IMAGE_END|>
+                {%- elif content_item.type in ['video_url', 'video'] -%}
+                    {%- set video_count.value = video_count.value + 1 -%}
+                    Video {{ video_count.value }}:<|VIDEO_START|><|VIDEO_PLACEHOLDER|><|VIDEO_END|>
+                {%- endif -%}
+            {%- endfor -%}
+        {%- endif -%}
+        {%- if message.role == 'system' -%}
+            {{- '
+            ' -}}
+        {%- endif -%}
+    {%- elif message.role == 'assistant' -%}
+        {%- macro extract_text_content(content_field) -%}
+            {%- if content_field is string -%}
+                {{- content_field -}}
+            {%- elif content_field is iterable and content_field is not string -%}
+                {%- set ns = namespace(text_parts=[]) -%}
+                {%- set text_parts = [] -%}
+                {%- for item in content_field -%}
+                    {%- if item.type == 'text' -%}
+                        {%- set ns.text_parts = ns.text_parts + [item.text] -%}
+                    {%- endif -%}
+                {%- endfor -%}
+                {{- ns.text_parts | join("") -}}
+            {%- else -%}
+                {{- '' -}}
+            {%- endif -%}
+        {%- endmacro -%}
+        {%- set reasoning_content = extract_text_content(message.reasoning_content) -%}
+        {%- set content = extract_text_content(message.content) -%}
+        {%- if '</think>' in content %}
+            {%- set reasoning_content = content.split('</think>')[0].rstrip('
+                        ').split('<think>')[-1].lstrip('
+                        ') %}
+            {%- set content = content.split('</think>')[-1].lstrip('
+                        ') %}
+        {%- endif %}
+        {%- if reasoning_content %}
+            {{- '
+            ' + 'Assistant: ' + '<think>
+            ' + reasoning_content.strip('
+                        ') + '
+            </think>
+            ' + content.lstrip('
+            ') }}
+        {%- else %}
+            {{- '
+            ' + 'Assistant: ' + content }}
+        {%- endif %}
+        {{- '<|end_of_sentence |>' }}
+    {%- endif -%}
+{%- endfor -%}
+{%- if add_generation_prompt is not defined or add_generation_prompt is true %}
+    {{- '
+Assistant: ' -}}
+    {%- if (enable_thinking is defined and enable_thinking is false) or enable_thinking is not defined %}
+        {{- '<think>
+
+</think>
+
+' }}
+    {%- endif %}
+    {%- if enable_thinking is defined and enable_thinking is true %}{{- '<think>' }}{%- endif %}
+{%- endif %}

config.json

@@ -0,0 +1,72 @@
+{
+  "architectures": [
+    "Ernie4_5_VLMoeForConditionalGeneration"
+  ],
+  "dtype": "float32",
+  "image_end_token_id": 6,
+  "image_start_token_id": 5,
+  "image_token_id": 7,
+  "model_type": "ernie4_5_vl_moe",
+  "text_config": {
+    "bos_token_id": null,
+    "eos_token_id": null,
+    "hidden_act": "silu",
+    "hidden_size": 16,
+    "initializer_range": 0.02,
+    "intermediate_size": 32,
+    "max_position_embeddings": 512,
+    "mlp_layer_types": [
+      "dense",
+      "sparse"
+    ],
+    "model_type": "ernie4_5_vl_moe_text",
+    "moe_intermediate_size": [
+      32,
+      32
+    ],
+    "moe_k": 2,
+    "moe_norm_min": 1e-12,
+    "moe_num_experts": 8,
+    "moe_num_shared_experts": 2,
+    "num_attention_heads": 2,
+    "num_hidden_layers": 2,
+    "num_key_value_heads": 1,
+    "output_router_logits": false,
+    "pad_token_id": null,
+    "rms_norm_eps": 1e-05,
+    "rope_parameters": {
+      "mrope_section": [
+        1,
+        1,
+        2
+      ],
+      "rope_theta": 500000.0,
+      "rope_type": "default",
+      "type": "default"
+    },
+    "router_aux_loss_coef": 0.001,
+    "tie_word_embeddings": true,
+    "use_bias": false,
+    "use_cache": true,
+    "vocab_size": 103368
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.16.0.dev0",
+  "video_end_token_id": 4,
+  "video_start_token_id": 3,
+  "video_token_id": 8,
+  "vision_config": {
+    "depth": 2,
+    "hidden_act": "silu",
+    "hidden_size": 32,
+    "in_channels": 3,
+    "initializer_range": 0.02,
+    "intermediate_size": 32,
+    "model_type": "ernie4_5_vl_moe_vision",
+    "num_heads": 2,
+    "patch_size": 14,
+    "rms_norm_eps": 1e-06,
+    "spatial_merge_size": 1,
+    "temporal_merge_size": 2
+  }
+}

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "transformers_version": "5.16.0.dev0",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d115a74a1c941ed9ea70abca932951581d1dd1568e40d8f7d75ec6ea14ece78
+size 6903104

preprocessor_config.json

@@ -0,0 +1,27 @@
+{
+  "data_format": "channels_first",
+  "do_convert_rgb": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_processor_type": "Ernie4_5_VL_MoeImageProcessor",
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "merge_size": 2,
+  "patch_size": 14,
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "longest_edge": 4842768,
+    "shortest_edge": 3136
+  },
+  "temporal_patch_size": 1
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94718bd0ed3c1f3591a1cf2c517d6364283457f2fd7e95ae4d201138e922079d
+size 11590466

tokenizer_config.json

@@ -0,0 +1,51 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "<|begin_of_sentence|>",
+  "eos_token": "</s>",
+  "extra_special_tokens": [
+    "<|IMAGE_PLACEHOLDER|>",
+    "<|IMAGE_START|>",
+    "<|IMAGE_END|>",
+    "<|VIDEO_PLACEHOLDER|>",
+    "<|VIDEO_START|>",
+    "<|VIDEO_END|>",
+    "<think>",
+    "</think>"
+  ],
+  "header_end_token": "<mask:7>",
+  "header_start_token": "<mask:6>",
+  "image_end_token": "<|IMAGE_END|>",
+  "image_start_token": "<|IMAGE_START|>",
+  "image_token": "<|IMAGE_PLACEHOLDER|>",
+  "is_local": true,
+  "local_files_only": false,
+  "mask_token": "<mask:1>",
+  "model_max_length": 131072,
+  "model_specific_special_tokens": {
+    "header_end_token": "<mask:7>",
+    "header_start_token": "<mask:6>",
+    "image_end_token": "<|IMAGE_END|>",
+    "image_start_token": "<|IMAGE_START|>",
+    "image_token": "<|IMAGE_PLACEHOLDER|>",
+    "sys_end_token": "<mask:5>",
+    "sys_start_token": "<mask:4>",
+    "video_end_token": "<|VIDEO_END|>",
+    "video_start_token": "<|VIDEO_START|>",
+    "video_token": "<|VIDEO_PLACEHOLDER|>"
+  },
+  "pad_token": "<unk>",
+  "padding_side": "left",
+  "processor_class": "Ernie4_5_VL_MoeProcessor",
+  "sep_token": "<|end_of_sentence|>",
+  "sys_end_token": "<mask:5>",
+  "sys_start_token": "<mask:4>",
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false,
+  "video_end_token": "<|VIDEO_END|>",
+  "video_start_token": "<|VIDEO_START|>",
+  "video_token": "<|VIDEO_PLACEHOLDER|>"
+}