[RFC] Sharded embeddings in separate FSDP group #182
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Outdated suggestions cannot be applied.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.
Stack from ghstack (oldest at bottom):
If we shard the embeddings as a separate FSDP parameter group, then:
This saves ~the embedding parameter size from peak memory without any decrease to WPS on the first order. (It introduces extra 2 all-gathers and 1 reduce-scatter, which can be bad for communication latency at large scale.)
For example, for Llama-7B with bf16 mixed precision, we save ~0.84 GiB, and on 8 GPUs, there is no noticeable effect on MFU.