[PR #278] [CLOSED] Support another low-bit format for low-latency combine #499

New issue

Closed

opened 2026-10-27 21:03:15 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-27 21:03:15 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/278
Author: @LyricZhao
Created: 7/4/2025
Status: ❌ Closed

Base: main ← Head: logfmt

📝 Commits (10+)

7930110 feat: logfmt
b8e046d slow
fac6ebf fast
b9805d7 consec
34fd9d4 t
4f564aa no logfmt?
58eb2cf pack
2a67a8d need clean
f772b6a one kernel
c5dc0be direct fuse

📊 Changes

8 files changed (+556 additions, -64 deletions)

View changed files

📝 csrc/config.hpp (+4 -1)
📝 csrc/deep_ep.cpp (+14 -2)
📝 csrc/deep_ep.hpp (+2 -2)
📝 csrc/kernels/api.cuh (+9 -0)
📝 csrc/kernels/internode_ll.cu (+451 -24)
📝 csrc/kernels/utils.cuh (+40 -6)
📝 deep_ep/buffer.py (+7 -7)
📝 tests/test_low_latency.py (+29 -22)

📄 Description

No description provided

_{🔄 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/278 **Author:** [@LyricZhao](https://github.com/LyricZhao) **Created:** 7/4/2025 **Status:** ❌ Closed **Base:** `main` ← **Head:** `logfmt` --- ### 📝 Commits (10+) - [`7930110`](https://github.com/deepseek-ai/DeepEP/commit/7930110460a3225e9b5273182ce87cc5f0950a14) feat: logfmt - [`b8e046d`](https://github.com/deepseek-ai/DeepEP/commit/b8e046d7a9acd6f0549bf84dc50590b00381988a) slow - [`fac6ebf`](https://github.com/deepseek-ai/DeepEP/commit/fac6ebf5a9a24ab9d1ae920796e9433ec938f1ba) fast - [`b9805d7`](https://github.com/deepseek-ai/DeepEP/commit/b9805d71906772b87909a8de1420467ec4c24310) consec - [`34fd9d4`](https://github.com/deepseek-ai/DeepEP/commit/34fd9d44c34ac57dffa5491a1ceb6487c9fe6246) t - [`4f564aa`](https://github.com/deepseek-ai/DeepEP/commit/4f564aa81c6e0204d046aa0ef3cada3d4f358340) no logfmt? - [`58eb2cf`](https://github.com/deepseek-ai/DeepEP/commit/58eb2cff41628249a34bd0231a6574d6d7876b57) pack - [`2a67a8d`](https://github.com/deepseek-ai/DeepEP/commit/2a67a8d56054b0804bb9d94e1f9d1d2f90a99618) need clean - [`f772b6a`](https://github.com/deepseek-ai/DeepEP/commit/f772b6a7ecac321b19480a922a8957bd72eb96ab) one kernel - [`c5dc0be`](https://github.com/deepseek-ai/DeepEP/commit/c5dc0be95955f7936cbf86f41cd15ae53f6bb746) direct fuse ### 📊 Changes **8 files changed** (+556 additions, -64 deletions) <details> <summary>View changed files</summary> 📝 `csrc/config.hpp` (+4 -1) 📝 `csrc/deep_ep.cpp` (+14 -2) 📝 `csrc/deep_ep.hpp` (+2 -2) 📝 `csrc/kernels/api.cuh` (+9 -0) 📝 `csrc/kernels/internode_ll.cu` (+451 -24) 📝 `csrc/kernels/utils.cuh` (+40 -6) 📝 `deep_ep/buffer.py` (+7 -7) 📝 `tests/test_low_latency.py` (+29 -22) </details> ### 📄 Description _No description provided_ --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-27 21:03:15 +00:00