Add high-priority A2A stream and HybridEP preprocessing SMs - #4694
Merged
Phlip79 merged 5 commits intoMay 17, 2026
Conversation
Port the main-applicable parts of PR NVIDIA#4401. - Add high_priority_a2a_comm_stream for combined 1F1B A2A overlap stream creation. - Add moe_hybridep_num_sms_preprocessing and plumb it through HybridEP dispatch initialization. The dev PR's overflow_flag handle-index change targets moe_expert_rank_capacity_factor code that is not present on main, so it is intentionally omitted here. Original dev commit: 7ae10f2.
gdengk
marked this pull request as ready for review
May 12, 2026 21:28
Victarry
approved these changes
May 13, 2026
yashaswikarnati
approved these changes
May 13, 2026
kvareddy
approved these changes
May 13, 2026
fanshiqing
approved these changes
May 14, 2026
ericharper
approved these changes
May 15, 2026
svcnvidia-nemo-ci
added
Approved
ericharper
enabled auto-merge
May 15, 2026 23:11
Contributor
|
/ok to test e050ac4 |
Member
|
/ok to test ba9ba5c |
Contributor
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/26002924178 |
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
May 17, 2026
Contributor
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/26004825546 |
Victarry
pushed a commit
to yanring/Megatron-LM
that referenced
this pull request
May 18, 2026
* origin/main: (138 commits) Refactor CUDA graph API: decompose cuda_graph_scope into full_iteration impl, inference scope, and per-layer capture modules (NVIDIA#4292) Add high-priority A2A stream and HybridEP preprocessing SMs (NVIDIA#4694) add is_torch_min_version in fsdp src (NVIDIA#4812) [Main][feat] Support A2A Overlap for Megatron-FSDP (NVIDIA#3797) Reorder mtp_post_process after attention backward in 1F1B schedule plan (NVIDIA#4695) [fix] Use MSC for checking checkpoint existence (NVIDIA#4251) Combine GEMM + SwiGLU fused MLP PRs (3890, 4071, 4095, 4219, 4311, 4324) → main (NVIDIA#4636) Strengthen test_checkpoint to verify distributed checkpoint behavior (NVIDIA#4711) Disable MSC by default; opt in via --enable-msc (NVIDIA#4629) additional tests for nvrx (NVIDIA#4522) Update copy-pr-bot.yaml [skip ci] ci: tolerate git-gc race in /home/runner chown after checkout (NVIDIA#4808) Inference: Optimize Prefill Engine Steps for Nemotron (NVIDIA#4764) chore: Update nightly tests golden values (NVIDIA#4805) ci: Update workflow to use same commit for building docker image and running tests (NVIDIA#4787) Update owners (NVIDIA#4794) Bump nvidia-modelopt>=0.44.0 (NVIDIA#4803) fix tokenizers in respect to newer transformers (NVIDIA#4608) Use Protocols to type-check linear_proj submodules of Attention (NVIDIA#3434) Fix recompute checkpointing + training CGs (NVIDIA#3919) ... # Conflicts: # megatron/core/transformer/moe/moe_utils.py
svcnvidia-nemo-ci
pushed a commit
to dimapihtar/Megatron-LM
that referenced
this pull request
Aug 4, 2026
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.
Ports the main-applicable portions of #4401.
Included:
The dev PR overflow_flag handle-index change targets moe_expert_rank_capacity_factor code that is not present on main, so it is intentionally omitted here.
Original dev PR: #4401
Original dev commit: 7ae10f2