Tim Vink

+1 for this! A potential route might to have `mlflow.models.infer_pip_requirements()` detect the presence of a `uv.lock` file. If present, it can determine the pip requirements by using a subprocess that...

I'm not sure how it works exactly, but things like databricks serving endpoints seems to build the environment from the conda specification. The way we work now is to export...

project_root parameter makes a lot of sense, good abstraction because it can be extended to any python environment system. But it does make it harder to test, as you'll have...

I would really appreciate additional optimizations to reduce inference latency. Basic models easily hit 100ms, which is very slow for many use cases. Additionally, for validating JSON data using pydantic...