Add mm processor args for Qwen3-VL

[CUHKSZzxy]

Motivation

User should be able to configure mm processor parameters for Qwen3-VL, e.g. min_pixels, max_pixels etc.
Therefore, we add a general mm processor kwargs argument for Qwen3-VL and possibly more VLMs.

https://github.com/QwenLM/Qwen3-VL?tab=readme-ov-file#new-qwen-vl-utils-usage

Usage

Adding mm_processor_kwargs into the message

curl http://0.0.0.0:23333/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Qwen/Qwen3-VL-8B-Instruct",
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "Can you describe these images?"
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg"
                    }
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg"
                    }
                }
            ]
        }
    ],
    "add_vision_id": true,
    "mm_processor_kwargs": {
        "min_pixels": 50176,
        "max_pixels": 50176
    }
  }'

Others

Remove deprecated env DEEPEP_MAX_BATCH_SIZE, refers to

• Sh/update deepep support DeepLink-org/DLBlas#86

[lvhan028]

should we add this option to "/generate" endpoint? cc @irexyc @RunningLeon

[irexyc]

This changed the number of arguments for self.model.preprocess, will it affect other models like internvl? https://github.com/CUHKSZzxy/lmdeploy/blob/44f0942078bfb5a98da12743f4529f4492636029/lmdeploy/vl/model/internvl.py#L193

[lvhan028]

I think it does. pr-test CI failed

[CUHKSZzxy]

should we add mm_processor_kwargs for all the rest vision models? or any other better solution

[lvhan028]

Let's address the internvl and qwen-vl series model first

[CUHKSZzxy]

Let's address the internvl and qwen-vl series model first

Have resolved this, by first checking if self.model.preprocess accpets mm_processor_kwargs or not. See the updated code.