[Bug]: Docker image for 0.5.4 does not include package timm==0.9.10 to run MiniCPMV

[bjornjee]

Your current environment

nil

🐛 Describe the bug

Running docker image 0.5.4 with the following entrypoint:

python3 -m vllm.entrypoints.openai.api_server --port 8088 --load-format auto --gpu-memory-utilization 0.90 --enforce-eager --disable-log-requests --model HwwwH/MiniCPM-V-2 --served-model-name HwwwH/MiniCPM-V-2 --trust-remote-code --max-model-len 4096 --tensor-parallel-size 1

Server errors out with error, code ref:

install timm==0.9.10

Willing to create MR for fix

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.

[jeejeelee]

timm needs to be installed locally , rather than as a common module.

[bjornjee]

curious if there is a reason why it is not baked in the Dockerfile. For users who run docker image directly, it will be better if the docker image has all required dependencies

[jeejeelee]

curious if there is a reason why it is not baked in the Dockerfile. For users who run docker image directly, it will be better if the docker image has all required dependencies

This module is already included in https://github.com/vllm-project/vllm/blob/main/requirements-test.txt.
For Dockerfile, if I understand correctly, it should not need to be manually installed again, see: https://github.com/vllm-project/vllm/blob/main/Dockerfile#L124

[DarkLight1337]

curious if there is a reason why it is not baked in the Dockerfile. For users who run docker image directly, it will be better if the docker image has all required dependencies

Different models may have their own dependencies, with some defined on the HuggingFace repo and thus outside of our control. It would be quite inefficient to install the dependencies for every model when you will only use a couple of them.

In any case, it should be easy to create a Dockerfile that adds an extra step on top of the existing one to install the dependency.