Include model context window limit on the API endpoint?

Nobody has claimed this yet.

I am developing a document summary application in which the user can dynamically select various model endpoints from `fastchat.serve.controller`.

Since document chunking depends entirely on the available context window of the model, it would be good if this parameter (e.g. something like `max_position_embeddings` for the vicuna-7b-v1.3 model) could be exposed on the API `/models` endpoint.

On a related note:

`curl http://localhost:8000/v1/models`

returns useful information, e.g.

`{"object":"list","data":[{"id":"vicuna-7b-v1.3","object":"model","created":1688626343,"owned_by":"fastchat","root":"vicuna-7b-v1.3","parent":null,"permission":[{"id":"modelperm-3Ki2CyP8sjBPGfZbtDDYqK","object":"model_permission","created":1688626343,"allow_create_engine":false,"allow_sampling":true,"allow_logprobs":true,"allow_search_indices":true,"allow_view":true,"allow_fine_tuning":false,"organization":"*","group":null,"is_blocking":false}]}]}`

but, e.g.,

`curl http://localhost:8000/v1/models/vicuna-7b-v1.3`

returns

`{"detail":"Not Found"}`

Is this a bug? According to the OpenAI API specification, this endpoint should return specific model parameters.

Start in fastchat.serve.controller and inspect the existing /v1/models route and its response shape. Compare the requested model-specific endpoint and context-window metadata with the OpenAI API specification; done means the model information is exposed consistently and the shown curl requests return the expected details.

Written by the indexing model from the issue text.