Fix Llama-70B-FSDP model loading issue #752
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Outdated suggestions cannot be applied.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.
When running large models with FSDP, if we try to load full model params to each card we will run out of CPU RAM (refer to this these PRs huggingface/transformers#25107 huggingface/accelerate#1777. This is avoided by loading model params only to card 0 and broadcasting params to other cards from card 0. Enabled CPU_RAM_EFFICIENT_LOADING configuration in accelerate to support above described functionality.