[PR #230] [CLOSED] Support next_n = 4 for SM100 MQA logits #301

New issue

Closed

opened 2026-10-26 21:15:59 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:15:59 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/230
Author: @RayWang96
Created: 11/19/2025
Status: ❌ Closed

Base: main ← Head: sm100_next_n_4

📝 Commits (2)

fdf1622 Fix SM90 MQA logits
b6ccd44 Add next_n = 4 support for sm100_fp8_paged_mqa_logits

📊 Changes

6 files changed (+50 additions, -28 deletions)

View changed files

📝 csrc/apis/attention.hpp (+8 -3)
📝 csrc/jit_kernels/impls/smxx_fp8_mqa_logits.hpp (+1 -1)
📝 csrc/jit_kernels/impls/smxx_fp8_paged_mqa_logits.hpp (+15 -7)
📝 deep_gemm/include/deep_gemm/impls/sm100_fp8_mqa_logits.cuh (+1 -1)
📝 deep_gemm/include/deep_gemm/impls/sm100_fp8_paged_mqa_logits.cuh (+20 -14)
📝 tests/test_attention.py (+5 -2)

📄 Description

Use a cluster to process a query instead of only using an SM.

_{🔄 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/DeepGEMM/pull/230 **Author:** [@RayWang96](https://github.com/RayWang96) **Created:** 11/19/2025 **Status:** ❌ Closed **Base:** `main` ← **Head:** `sm100_next_n_4` --- ### 📝 Commits (2) - [`fdf1622`](https://github.com/deepseek-ai/DeepGEMM/commit/fdf16223e1741badb53bff67362eb2e08af5258d) Fix SM90 MQA logits - [`b6ccd44`](https://github.com/deepseek-ai/DeepGEMM/commit/b6ccd44c043471f39e4f68c15e9291b0e5d5c471) Add next_n = 4 support for sm100_fp8_paged_mqa_logits ### 📊 Changes **6 files changed** (+50 additions, -28 deletions) <details> <summary>View changed files</summary> 📝 `csrc/apis/attention.hpp` (+8 -3) 📝 `csrc/jit_kernels/impls/smxx_fp8_mqa_logits.hpp` (+1 -1) 📝 `csrc/jit_kernels/impls/smxx_fp8_paged_mqa_logits.hpp` (+15 -7) 📝 `deep_gemm/include/deep_gemm/impls/sm100_fp8_mqa_logits.cuh` (+1 -1) 📝 `deep_gemm/include/deep_gemm/impls/sm100_fp8_paged_mqa_logits.cuh` (+20 -14) 📝 `tests/test_attention.py` (+5 -2) </details> ### 📄 Description Use a cluster to process a query instead of only using an SM. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 21:15:59 +00:00