[PR #654] [CLOSED] Change release scope of local-rank Rail bypass for dispatch #690

New issue

Closed

opened 2026-10-27 21:05:17 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-27 21:05:17 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/654
Author: @huakainomore
Created: 6/1/2026
Status: ❌ Closed

Base: main ← Head: main

📝 Commits (1)

dd1c1d1 fix: change release scope from sys to gpu for dispatch update tail

📊 Changes

3 files changed (+33 additions, -1 deletions)

View changed files

📝 deep_ep/include/deep_ep/common/handle.cuh (+24 -0)
📝 deep_ep/include/deep_ep/common/ptx.cuh (+8 -0)
📝 deep_ep/include/deep_ep/impls/hybrid_dispatch.cuh (+1 -1)

📄 Description

~30% improvement when ep16 with 12 SMs
5% improvement when ep16 with 16 SMs

when use less sms or small EP size(like EP 16), this will improve performance.
releated to issue 647

_{🔄 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/654 **Author:** [@huakainomore](https://github.com/huakainomore) **Created:** 6/1/2026 **Status:** ❌ Closed **Base:** `main` ← **Head:** `main` --- ### 📝 Commits (1) - [`dd1c1d1`](https://github.com/deepseek-ai/DeepEP/commit/dd1c1d1ed2bb2bff34927d7d3bd5a72069eba94a) fix: change release scope from sys to gpu for dispatch update tail ### 📊 Changes **3 files changed** (+33 additions, -1 deletions) <details> <summary>View changed files</summary> 📝 `deep_ep/include/deep_ep/common/handle.cuh` (+24 -0) 📝 `deep_ep/include/deep_ep/common/ptx.cuh` (+8 -0) 📝 `deep_ep/include/deep_ep/impls/hybrid_dispatch.cuh` (+1 -1) </details> ### 📄 Description 1. ~30% improvement when ep16 with 12 SMs 2. 5% improvement when ep16 with 16 SMs when use less sms or small EP size(like EP 16), this will improve performance. releated to [issue 647](https://github.com/deepseek-ai/DeepEP/issues/647) --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-27 21:05:17 +00:00