Fix Sparsity Logs on FSDP Model Save #2203
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Model sparsity was not being logged correctly on FSDP save. This was because when we save an FSDP model we gather it onto the CPU in the main process and call save_pretrained on the wrapped model.
This results in the wrapped model being passed to the global sparsity function with incomplete information. We need the full state dict as well to properly calcualte the sparsity. I also add some other log cleanup so we aren't printing out as much sparsity information between train stages
Example
Launch with FSDP:
accelerate launch --config_file integrations/huggingface-transformers/finetuning/example_fsdp_config.yaml test.py
Output:
All the parameters get picked up, previously only the 23 wrapped transformers layers were showing up as parameters