[MoE] Improvement of shared expert overlap, support shared expert overlap for FlexDispatcher #2207
+214
−121
Conversation
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
1. Add shared expert overlap for FlexDispatcher 2. Add stream wait for cases where CUDA_DEVICE_MAX_CONNECTIONS> 1 to prevent shared expert GEMM launched too early. 3. Change fc1 location of shared experts in A2A dispatcher for better overlap.
Contributor
|
@fanshiqing can you please take a look at this MR? |
yanring
added
the
Expert Review
Skylion007
reviewed
Nov 11, 2025
|
|
||
| if self.stream is None: | ||
| self.stream = torch.cuda.Stream() | ||
| if SharedExpertMLP.stream is None: |
Contributor
There was a problem hiding this comment.
Yikes, isn't this a ClassVar now? Will we never need 2 different stream for the same class?
Contributor
Author
There was a problem hiding this comment.
Yeah. It's intended. There will not be two shared experts overlap with each other.
Moreover, if we creating a new stream for each instance, PyTorch may run out of streams from stream pool and reuse existing stream. This may cause interference and unwanted behavior.
Contributor
|
/ok to test 3e308ac |
pablo-garay
approved these changes
Nov 21, 2025
kvareddy
reviewed
Nov 22, 2025
| group=group, | ||
| ) | ||
| if use_nccl_stream: | ||
| handle = torch.distributed.all_to_all_single( |
Contributor
There was a problem hiding this comment.
What is the differene between if and else paths?
Contributor
|
@fanshiqing can you please take a look at this MR? |
yanring
changed the title
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 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Outdated suggestions cannot be applied.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 ?
Design doc: https://docs.google.com/document/d/1whtnUiw1hpfdkjFss_g5P8fyIT9xBA5XJvmdklFes48/edit?usp=sharing
Changelog:
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
[email protected]or[email protected].Merging your PR
Any member of core-adlr and
core-nemowill be able to merge your PR.