[PR #637] [MERGED] Merge some update to main #677

Closed
opened 2026-10-27 21:05:11 +00:00 by bsakaguchi · 0 comments
Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/637
Author: @sphish
Created: 5/20/2026
Status: ✅ Merged
Merged: 5/21/2026
Merged by: @sphish

Base: main ← Head: update-main


📝 Commits (1)

  • acfbae8 Cherry-pick origin/main 6d1e8fd..0534254 onto update-main

📊 Changes

17 files changed (+618 additions, -133 deletions)

View changed files

📝 csrc/elastic/buffer.hpp (+65 -29)
📝 csrc/indexing/main.cu (+1 -0)
📝 csrc/kernels/backend/api.cuh (+10 -3)
📝 csrc/kernels/backend/nccl.cu (+16 -14)
➕ csrc/kernels/backend/symmetric.hpp (+322 -0)
📝 csrc/kernels/elastic/engram.hpp (+46 -13)
📝 csrc/utils/lazy_driver.hpp (+2 -0)
📝 deep_ep/buffers/elastic.py (+62 -17)
📝 deep_ep/include/deep_ep/common/handle.cuh (+6 -0)
📝 deep_ep/include/deep_ep/common/layout.cuh (+1 -2)
📝 deep_ep/include/deep_ep/impls/combine_utils.cuh (+3 -3)
📝 deep_ep/include/deep_ep/impls/engram_fetch.cuh (+29 -41)
➕ deep_ep/include/deep_ep/impls/engram_fetch_wait.cuh (+31 -0)
📝 deep_ep/include/deep_ep/impls/hybrid_combine.cuh (+6 -2)
📝 deep_ep/utils/comm.py (+1 -1)
📝 tests/elastic/test_agrs.py (+12 -3)
📝 tests/elastic/test_engram.py (+5 -5)

📄 Description

What's New

Bug Fixes

  • Fix a rare numerical error in multi-node hybrid combine

Features

  • Add CPU buffer support for Engram
  • Remove alignment requirement for all-gather

Improvements

  • Reuse torch NCCL comm by default to reduce GPU memory usage
  • Improve combine/combine-epilogue performance for certain cases

🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/DeepEP/pull/637 **Author:** [@sphish](https://github.com/sphish) **Created:** 5/20/2026 **Status:** ✅ Merged **Merged:** 5/21/2026 **Merged by:** [@sphish](https://github.com/sphish) **Base:** `main` ← **Head:** `update-main` --- ### 📝 Commits (1) - [`acfbae8`](https://github.com/deepseek-ai/DeepEP/commit/acfbae86439f2c4a6381ec18149753b05dcebee8) Cherry-pick origin/main 6d1e8fd..0534254 onto update-main ### 📊 Changes **17 files changed** (+618 additions, -133 deletions) <details> <summary>View changed files</summary> 📝 `csrc/elastic/buffer.hpp` (+65 -29) 📝 `csrc/indexing/main.cu` (+1 -0) 📝 `csrc/kernels/backend/api.cuh` (+10 -3) 📝 `csrc/kernels/backend/nccl.cu` (+16 -14) ➕ `csrc/kernels/backend/symmetric.hpp` (+322 -0) 📝 `csrc/kernels/elastic/engram.hpp` (+46 -13) 📝 `csrc/utils/lazy_driver.hpp` (+2 -0) 📝 `deep_ep/buffers/elastic.py` (+62 -17) 📝 `deep_ep/include/deep_ep/common/handle.cuh` (+6 -0) 📝 `deep_ep/include/deep_ep/common/layout.cuh` (+1 -2) 📝 `deep_ep/include/deep_ep/impls/combine_utils.cuh` (+3 -3) 📝 `deep_ep/include/deep_ep/impls/engram_fetch.cuh` (+29 -41) ➕ `deep_ep/include/deep_ep/impls/engram_fetch_wait.cuh` (+31 -0) 📝 `deep_ep/include/deep_ep/impls/hybrid_combine.cuh` (+6 -2) 📝 `deep_ep/utils/comm.py` (+1 -1) 📝 `tests/elastic/test_agrs.py` (+12 -3) 📝 `tests/elastic/test_engram.py` (+5 -5) </details> ### 📄 Description ## What's New **Bug Fixes** - Fix a rare numerical error in multi-node hybrid combine **Features** - Add CPU buffer support for Engram - Remove alignment requirement for all-gather **Improvements** - Reuse torch NCCL comm by default to reduce GPU memory usage - Improve combine/combine-epilogue performance for certain cases --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
bsakaguchi 2026-10-27 21:05:11 +00:00
Sign in to join this conversation.
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepEP#677
No description provided.