I am using vllm to start a video processing service for qwen2vl-7b, but there is an error message when passing in video_url. The error message is Unknown part type video_url, and the vllm version is 0.6.3.post1[Usage]:

[hyyuananran]

Your current environment

The output of `python collect_env.py`

How would you like to use vllm

I want to run inference of a [specific model](put link here). I don't know how to integrate it with vllm.

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[DarkLight1337]

Please show the command that you used to serve the model, and also how you sent the request to the server.

[hyyuananran]

python -m vllm.entrypoints.openai.api_server --host 0.0.0.0 --port 8088 --model /app/qwen2vl-7b/ --dtype auto --tensor-parallel-size 1 --gpu-memory-utilization 0.95 -max-model-len 8096 --served-model-name qwen2vl-7b --trust-remote-code

query = {"model":"qwen2vl-7b",
"messages":[
{"role":"user",
"content":[
{"type":"text","text":"xxxxxxxxx"},
{"type":"video_url","video_url":{"url":xxxxxxxxx}
}]
}]
}

[DarkLight1337]

What type of URL are you passing in? In v0.6.3, only data:video/jpeg URLs are supported. v0.6.6 supports more types of data URLs, while #11620 enables file URLs as well.

[hyyuananran]

It is a URL for an S3 repository, and the file format is mp4