MLFlow logging using PyTorch Lightning, how to log by steps instead of epochs?

[FeryET]

Hi

Reading MLFlow's pytorch integration, there is an autolog option using which you can log using mlflow easily. But there arises some problems:

1 - I cannot apparently log by steps, it can only log the metric after N epochs (N is a natural number).
2 - I cannot log the trainer's parameters. Things like grad_clip_value, or early stopping patience and etc. This is extremely important in hyperparameter optimization.

How can I achieve these? Thank you.

P.S: Overall Lightning + MLFlow need more documentation, both are very good tools and apparently they integrate well enough, but the documnetation is lacking a bit. I know this is a community endeavor, but still I find it a bit behind the curve.