Mlflow log keras model with strategy MultiWorkerMirroredStrategy

Ray Libraries (Data, Train, Tune, Serve) Ray Train
1

What is the best practice for logging tf.keras models using a combination of mlflow, Ray train, Ray tune and MultiWorkerMirroredStrategy? Seems there is the mlflow callback for Ray train, but this only helps with parameters and metrics, not the model part.

How severe does this issue affect your experience of using Ray?

  • Medium: It contributes to significant difficulty to complete my task, but I can work around it.
2

Hi @jharaldson, you could look into using the Mlflow mixin (Using MLflow with Tune — Ray 1.11.0, also exists for Ray Train). Then you can use the regular mlflow python API to e.g. create a model version in your mlflow artifact repository.

1 Like