[DEV][NVFP4] Fix NVFP4 Selective Activation Recompute - #2036
Merged
Conversation
6 tasks
hxbai
approved these changes
Oct 30, 2025
yanring
enabled auto-merge
October 30, 2025 01:18
yanring
approved these changes
Oct 30, 2025
Contributor
|
/ok to test a982793 |
@yanring, there was an error processing your request: See the following link for more information: https://docs.gha-runners.nvidia.com/cpr/e/2/ |
Contributor
|
/ok to test 7598039 |
zhongbozhu
force-pushed
the
zhongbo/fix_nvfp4_act_recomp
branch
from
November 3, 2025 18:14
7598039 to
0581182
Compare
Member
|
/ok to test dc682de |
qijiaxing
reviewed
Nov 6, 2025
| return get_fp8_align_size(config.fp8_recipe) | ||
| elif config.fp4: | ||
| return get_fp4_align_size(config.fp4_recipe) | ||
| return 16 |
Contributor
Author
There was a problem hiding this comment.
This 16 is for other fp8 recipes like per-tensor current scaling.
BF16 shouldn't call this function.
Contributor
Author
|
Bumping up align size to 64 for future grouped nvfp4 kernels: NVIDIA/TransformerEngine#2351 |
17 tasks
Contributor
|
/ok to test de9127b |
48 tasks
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
8 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 ?
NVFP4 MOE is already available in dev branch, but it has some issues with selective activation recompute because some logic hardcoded
FP8. This PR is to fix it.PR to
mainbranch also has this fix: #1985This PR also bumps up align size of FP4 to 64 from 32 based on TE PR: NVIDIA/TransformerEngine#2351
Contribution process
flowchart LR A[Pre-checks] --> B[PR Tests] subgraph Code Review/Approval C1[Expert Review] --> C2[Final Review] end B --> C1 C2 --> D[Merge]Pre-checks
Core 0.8)Code review
The following process is enforced via the CODEOWNERS file for changes into
megatron/core. For changes outside ofmegatron/core, it is up to the PR author whether or not to tag the Final Reviewer team.For MRs into `main` branch
(Step 1): Add PR label
Expert Review(Step 2): Collect the expert reviewers reviews
Expert Reviewlabel when your PR is ready for review.Final Review might get declined if these requirements are not fulfilled.
(Step 3): Final Review
Final Reviewlabel(Optional Step 4): Cherry-pick into release branch
If this PR also needs to be merged into
core_r*release branches, after this PR has been merged, selectCherry-pickto open a new PR into the release branch.For MRs into `dev` branch
The proposed review process for `dev` branch is under active discussion.MRs are mergable after one approval by either
eharper@nvidia.comorzijiey@nvidia.com.Merging your PR
Any member of core-adlr and
core-nemowill be able to merge your PR.