feat(refit): support BF16 to MXFP8 NCCL reshard - #3477
Open
seonjinn wants to merge 4 commits into
Open
Conversation
Contributor
Author
|
/ok to test 83010d8 |
Signed-off-by: seonjinn <sna@nvidia.com>
seonjinn
force-pushed
the
sna/nccl-reshard-bf16-mxfp8
branch
from
August 3, 2026 19:56
83010d8 to
f3dcd2a
Compare
Contributor
Author
|
/ok to test f3dcd2a |
Signed-off-by: seonjinn <sna@nvidia.com>
Contributor
Author
|
/ok to test 4d1522b |
Signed-off-by: seonjinn <sna@nvidia.com>
Contributor
Author
|
/ok to test d63ed16 |
Contributor
Author
|
/ok to test 6f57c1b799bf980e6a99bc9dc530ce190029e7df |
@seonjinn, there was an error processing your request: See the following link for more information: https://docs.gha-runners.nvidia.com/cpr/e/2/ |
Contributor
Author
|
/ok to test 6f57c1b |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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?
This PR extends NCCL Reshard refit to support BF16 trainer storage with an MXFP8 vLLM rollout engine.
This PR does not add trainer-side prequantization or alter the MoE layout transform. The independent batched-shuffle optimization is in #3478.
Why?
NCCL Reshard previously required matching trainer and rollout storage. A BF16-training/MXFP8-rollout recipe therefore failed validation even though the destination worker can quantize its already-resharded local shard without changing the collective protocol.
Correctness