[Question] Is it necessary to adapt report and Checkpointing to the newly introduced session and Checkpoint API of Ray AIR?

ray-project / ray_lightning

Pytorch Lightning Distributed Accelerators using Ray

Apache License 2.0

211 stars 34 forks source link

[Question] Is it necessary to adapt report and Checkpointing to the newly introduced session and Checkpoint API of Ray AIR? #243

Open MarkusSpanring opened 1 year ago

MarkusSpanring commented 1 year ago

As the title already suggests. I was looking into the latest examples of PyTorch with Ray Tune here and found that resuming from a checkpoint and reporting to tune is now handled via the session API provided by Ray AIR.

Is it necessary to adapt TuneReportCallback and TuneReportCheckpointCallback accordingly?