When I generate a client, it breaks because it fails to process the schema of ChatCompletionRequestMessage
These fix that:
- I think `Union[Literal["user"], Literal["channel"], ...]` is the same as Literal["user", "channel", ...]
- Turns out default value `Literal["user"]` isn't JSON serializable, so replace with "user"
`model` is ignored, but currently marked "optional"... on the one hand could mark "required" to make it explicit in case the server supports multiple llama's at the same time, but also could delete it since its ignored. decision: mark it required for the sake of openai api compatibility.
I think out of all parameters, `model` is probably the most important one for people to keep using even if its ignored for now.
Define an init_llama function that starts llama with supplied settings instead of just doing it in the global context of app.py
This allows the test to be less brittle by not needing to mess with os.environ, then importing the app
Change batch size to the llama.cpp default of 8. I've seen issues in llama.cpp where batch size affects quality of generations. (It shouldn't) But in case that's still an issue I changed to default.
Set auto-determined num of threads to 1/2 system count. ggml will sometimes lock cores at 100% while doing nothing. This is being addressed, but can cause bad experience for user if pegged at 100%
