Yuki Watanabe

(pretty strange why the test fails only in CI, investigating)

Things are way more complicated than I thought - giving up here and remove 3.1.0 label. Important but not blocking.

Happy to merge the PR once it passed all testing. At the same time, @redfave @marinegor Isn't the right fix updating Pytorch Lightning logger to pass string as an artifact...

@marinegor What I mean is fixing the pytorch logger implementation to pass string. https://github.com/Lightning-AI/pytorch-lightning/blob/e088694d411790747af1a52a59a20e6efaa1bebd/src/lightning/pytorch/loggers/mlflow.py#L364

Thank you for the feature request, @harshilprajapati96! Yes, chat-style support is on our roadmap for prompt registry. cc: @rohitarun-db

@harshilprajapati96 Thank you for the offering! Indeed it would be awesome if you can help implementing the feature. The proposed design looks solid, a few questions: > response_format: JSON schema...

@NevoleMarek The proposal totally makes sense! Would you go ahead and make a PR? I think we can add a new argument to `metrics.token_count` and move `_token_count_eval_fn` to an inner...

Hi, @kimminw00. The use case makes sense to me. We will investigate if we can extend the support for normal tracking experience.

@copilot I don't see any file diff. It seems you encountered some error before pushing the change.

@restato Thanks for the heads-up, resolved the blockers. This will be included in the MLflow 3.2 stable release next week.