Support model size calculation for FSDP & DeepSpeed (stage-3) #10291
Labels
feature
Is an improvement or enhancement
strategy: deepspeed
strategy: fsdp
Fully Sharded Data Parallel
🚀 Feature
Currently
LightningModule.model_size
(deprecated) relies on:https://github.com/PyTorchLightning/pytorch-lightning/blob/45c45dc7b018f9a2db60f5df1a3f7dbbb45ccb36/pytorch_lightning/utilities/memory.py#L165-L179
to calculate model size. Since we are going to keep
get_model_size_mb
utility function, we should have a way to calculate model_sizes for sharded models.cc @Borda @awaelchli @carmocca @SeanNaren @rohitgr7 @akihironitta
The text was updated successfully, but these errors were encountered: