Serve correctly on vLLM: fix image_processor_type, pre-fill latent block in chat template, pin vision_config

- preprocessor_config.json: image_processor_type Qwen2_5_VLImageProcessor -> Qwen2VLImageProcessor.
The old name is the transformers-fork class and is not in the auto map, so stock
transformers raised 'Unrecognized image processor'. The two classes produce
bit-identical pixel_values.
- chat_template: the add_generation_prompt branch now pre-fills the assistant turn with
<think> + 20 x <|latent_pad|> + </think>. Without it a served model receives zero latent
tokens and reverts to thinking in text (~11x the decode cost). Guarded by
add_generation_prompt, so training and lmms-eval render byte-identically to before.
Override per request with chat_template_kwargs {"num_latents": N}; 0 disables.
- config.json: vision_config written out explicitly rather than relying on transformers
class defaults. auto_map is kept.

chat_template.jinja

@@ -4,4 +4,5 @@ You are a helpful assistant.<|im_end|>
 {% if message['content'] is string %}{{ message['content'] }}<|im_end|>
 {% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>
 {% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant
-{% endif %}
+{% set n = num_latents if num_latents is defined else 20 %}{% if n > 0 %}<think>{{ '<|latent_pad|>' * n }}</think>
+{% endif %}{% endif %}

chat_template.json

@@ -1,3 +1,3 @@
 {
-  "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}"
-}
+  "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% set n = num_latents if num_latents is defined else 20 %}{% if n > 0 %}<think>{{ '<|latent_pad|>' * n }}</think>\n{% endif %}{% endif %}"
+}

config.json

@@ -48,14 +48,29 @@
   "use_sliding_window": false,
   "video_token_id": 151656,
   "vision_config": {
+    "depth": 32,
     "dtype": "bfloat16",
+    "fullatt_block_indexes": [
+      7,
+      15,
+      23,
+      31
+    ],
+    "hidden_act": "silu",
     "hidden_size": 1280,
     "in_chans": 3,
     "initializer_range": 0.02,
+    "intermediate_size": 3420,
     "model_type": "qwen2_5_vl",
+    "num_heads": 16,
+    "out_hidden_size": 3584,
+    "patch_size": 14,
+    "spatial_merge_size": 2,
     "spatial_patch_size": 14,
+    "temporal_patch_size": 2,
     "tokens_per_second": 2,
-    "torch_dtype": "bfloat16"
+    "torch_dtype": "bfloat16",
+    "window_size": 112
   },
   "vision_end_token_id": 151653,
   "vision_start_token_id": 151652,

preprocessor_config.json

@@ -8,7 +8,7 @@
     0.4578275,
     0.40821073
   ],
-  "image_processor_type": "Qwen2_5_VLImageProcessor",
+  "image_processor_type": "Qwen2VLImageProcessor",
   "image_std": [
     0.26862954,
     0.26130258,