[HTTP Server] Make model param optional in request

[youngkent]

We often have single model per deployment setup. Requiring model name in request create unnecessary coupling between service and client. This change make "model" param optional. When not given, it uses the first available model on the server to serve the request.

Test before the change, one need to send request like:

curl --insecure https://localhost:8000/v1/chat/completions \                                                                                                          
     -H "Content-Type: application/json" \
     -d '{
           "model": "/packages/8B_model",     
           "messages": [{"role": "user", "content": "Hello, vLLM!"}],
           "max_tokens": 100
         }' 

After the change, one can send the request like:

curl --insecure https://localhost:8000/v1/chat/completions \                                                                                                          
     -H "Content-Type: application/json" \
     -d '{ 
           "messages": [{"role": "user", "content": "Hello, vLLM!"}],
           "max_tokens": 100
         }' 

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[WoosukKwon]

@youngkent Thanks for submitting the PR!
It looks good to me, but maybe we need a second eye as I'm not very familiar with the frontend stuffs.

[simon-mo]

While this is breaking OpenAI compatibility/semantics, I actually think it's the right thing to do because vLLM only support one instance of the model anyway!

[simon-mo]

cc @DarkLight1337 @njhill

[simon-mo]

Please change one of the entrypoint test to reflect this behavior.

[njhill]

I'm pretty sure this was proposed in the past and decided against, but can't find the issue/PR now. In any case I don't remember why and I am also in favor of this.

[youngkent]

The huggingface file downloading error seems not relevant. Can I get a manual merge please? Thanks!
cc: @WoosukKwon @simon-mo @DarkLight1337 @njhill