Backwards-incompatible Changes
- openllm-core
Bump `attrs` to `23.2.0`
Added experimental helpers `.pydantic_model()` functions to convert current attrs-based class to its compatible pydantic class.
openllm
Updated OpenLLM architecture to new 1.2 BentoML.
`openllm.Runner` remains the old Runnable implementation. Therefore, if you still depends on the old architecture, make sure to use `openllm.Runner` instead of `llm.runner`.
`llm.runner` will now become an `bentoml.depends()` singleton, therefore, to avoid breaking change, make sure to set `OPENLLM_RUNNER_BEHAVIOUR=deprecated` in your environment variable. This is the default behaviour. To opt-in the new architecture, set `OPENLLM_RUNNER_BEHAVIOUR=new_impl`
[821](https://github.com/bentoml/openllm/issues/821)