Repository navigation
Conversation
`dp_size` says how many ranks see a distinct batch. FSDP2 is applied to a mesh of `dp_replicate_size * fsdp_size` ranks, which the two only coincide with while there is no context parallelism. Add `fsdp_mesh_size` for the second meaning and use it at every place that asked whether FSDP2 is applied. Values are unchanged.
Member
Author
|
Folded into the next PR: |
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What does this PR do?
Stack: #50493 → this PR →
DistributedConfig(cp_size=N)(next)dp_size(dp_replicate_size * fsdp_size) means two things today: the number of ranks that see different batches, and "FSDP2 is applied". With context parallelism, FSDP2 also shards over the ranks of a cp group, which share one batch, so the two differ.This adds
DistributedConfig.fsdp_mesh_sizeand uses it wherever "FSDP2 is applied" is meant (8 places inmixin.py, and the FSDP + PP check). Same values today: no behavior change.Tests
The existing FSDP / HSDP / EP tests pass on CPU (llama, qwen3_moe,
tests/utils/test_modeling_utils.py).