deepseek-ai/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP.git synced 2026-11-05 00:47:32 +00:00
Code Issues 308 Projects Packages Wiki Activity

[PR #295] [MERGED] third-party: Improvements to NVSHMEM Integration #507

New issue
Closed
opened 2026-10-27 21:03:22 +00:00 by bsakaguchi · 0 comments
bsakaguchi commented 2026-10-27 21:03:22 +00:00
Owner
Copy link

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/295
Author: @seth-howell
Created: 7/12/2025
Status: ✅ Merged
Merged: 7/16/2025
Merged by: @sphish

Base: main ← Head: main

📝 Commits (9)

  • 441833d third-party: Update tests to use upstream NVSHMEM
  • aa3187e third-party: Add CPU-assisted IBGDA support
  • 69f9dfe third-party: Update readme to reflect new features.
  • b79ca2b third-party: Add back nvshmem.patch.
  • 903711c setup.py: Clean up some extra prints.
  • 2a87339 setup.py: Add logic for detecting library locations from NVSHMEM wheels.
  • 35e1cd1 setup.py: Remove nvcc_dlink specific gencode
  • e6b4f52 buffer.py: Do not force the NIC handler to GPU.
  • c5d2202 third-party: Add link to blog post on CPU-Assisted IBGDA.

📊 Changes

5 files changed (+59 additions, -48 deletions)

View changed files

📝 csrc/kernels/ibgda_device.cuh (+12 -5)
📝 csrc/kernels/runtime.cu (+1 -0)
📝 deep_ep/buffer.py (+0 -1)
📝 setup.py (+24 -6)
📝 third-party/README.md (+22 -36)

📄 Description

  1. Use upstream NVSHMEM binaries when building DeepEP.
  2. Add back support for CPU-Assisted IBGDA.
  3. Remove the nvshmem host-side patch.

🔄 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/295 **Author:** [@seth-howell](https://github.com/seth-howell) **Created:** 7/12/2025 **Status:** ✅ Merged **Merged:** 7/16/2025 **Merged by:** [@sphish](https://github.com/sphish) **Base:** `main` ← **Head:** `main` --- ### 📝 Commits (9) - [`441833d`](https://github.com/deepseek-ai/DeepEP/commit/441833d3b1938146b0257be8ea1710e84d20f6d9) third-party: Update tests to use upstream NVSHMEM - [`aa3187e`](https://github.com/deepseek-ai/DeepEP/commit/aa3187efc0726bc1540b5258d5d9c784c30f5654) third-party: Add CPU-assisted IBGDA support - [`69f9dfe`](https://github.com/deepseek-ai/DeepEP/commit/69f9dfe20965596a28b59c5b241c4aca8ef35735) third-party: Update readme to reflect new features. - [`b79ca2b`](https://github.com/deepseek-ai/DeepEP/commit/b79ca2bbdfda75ba5bfbcd88f5d255f1ef92d519) third-party: Add back nvshmem.patch. - [`903711c`](https://github.com/deepseek-ai/DeepEP/commit/903711c6a25f18d7eecac4b057f177fb8f81cbc3) setup.py: Clean up some extra prints. - [`2a87339`](https://github.com/deepseek-ai/DeepEP/commit/2a873392ed0025a01006d6612569aa466afbbdf0) setup.py: Add logic for detecting library locations from NVSHMEM wheels. - [`35e1cd1`](https://github.com/deepseek-ai/DeepEP/commit/35e1cd1be19dc5b2affaa089675d425eab978fb7) setup.py: Remove nvcc_dlink specific gencode - [`e6b4f52`](https://github.com/deepseek-ai/DeepEP/commit/e6b4f527b2105dbbf017e7a8554933bef0706673) buffer.py: Do not force the NIC handler to GPU. - [`c5d2202`](https://github.com/deepseek-ai/DeepEP/commit/c5d220230ab4e9d826622b6401c7768c9dd4da56) third-party: Add link to blog post on CPU-Assisted IBGDA. ### 📊 Changes **5 files changed** (+59 additions, -48 deletions) <details> <summary>View changed files</summary> 📝 `csrc/kernels/ibgda_device.cuh` (+12 -5) 📝 `csrc/kernels/runtime.cu` (+1 -0) 📝 `deep_ep/buffer.py` (+0 -1) 📝 `setup.py` (+24 -6) 📝 `third-party/README.md` (+22 -36) </details> ### 📄 Description 1. Use upstream NVSHMEM binaries when building DeepEP. 2. Add back support for CPU-Assisted IBGDA. 3. Remove the nvshmem host-side patch. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
bsakaguchi 2026-10-27 21:03:22 +00:00
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
hybrid-ep
chore/remove-clang-format-yapf
v2.5
nvDev
try-fix-cu128
mori-ep
antgroup-opt
compat-old-torch
add-linter
eager-rdma
v1.2.1
Labels
Clear labels   
pull-request
Mirrored from GitHub Pull Request

No labels
pull-request
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
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#507
No description provided.