Can't get AsyncModel from extra-openai-models.yaml?

It looks like currently there is no way to get an AsyncModel from a custom model defined via extra-openai-models.yaml?

Looking at register_models in https://github.com/simonw/llm/blob/main/llm/default_plugins/openai_models.py , the standard OpenAI models all get both Chat and AsyncChat versions (or Completion, for gpt-3.5-instruct); however models defined in extra_models only get a Chat (or Completion) version, no Async.

It is not immediately clear to me whether it would be correct to set up an AsyncChat model version for all OpenAI chat completion API-compatible models, or whether there are some API-compatible models for which only a sync Chat model should be created. If there's a reason to only support synchronous calls for some API-compatible models, perhaps add an optional syncOnly: True flag in the yaml definition? Otherwise, it would be nice to create both Chat and AsyncChat versions for all extra OpenAI compatible models, as is currently done for the standard OpenAI models.

I'm using the python API to access several different models from multiple providers, including some locally-hosted models served through vllm's OpenAI compatible API. The best way I have found so far to access these vllm models is by defining them in extra-openai-models.yaml, but right now this seems to keep me restricted to using only the synchronous API.

Thanks much!

Start in llm/default_plugins/openai_models.py at register_models and compare how standard OpenAI models receive Chat and AsyncChat variants with how extra-openai-models.yaml entries are registered. Use the Python API examples in the issue to verify that custom OpenAI-compatible models expose the intended asynchronous access, while deciding whether synchronous-only models need an explicit configuration option.