[PR #307] [CLOSED] change sm100_fp8_mqa_logits to 2cta, and change mma acc output to f16 #353

New issue

Closed

opened 2026-10-26 21:16:36 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:16:36 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/307
Author: @benzh-2025
Created: 4/17/2026
Status: ❌ Closed

Base: nv_dev ← Head: benzh/2cta_with_half

📝 Commits (2)

f2d6a64 add sm100_fp8_mqa_logits_f16_weights
6ec4c1d change test_mqa_logits to avoid f16 overflow when sm100_fp8_mqa_logits_f16_weights

📊 Changes

7 files changed (+652 additions, -3 deletions)

View changed files

📝 csrc/apis/attention.hpp (+7 -1)
📝 csrc/indexing/main.cu (+1 -0)
📝 csrc/jit_kernels/impls/runtime_utils.hpp (+1 -0)
📝 csrc/jit_kernels/impls/smxx_fp8_fp4_mqa_logits.hpp (+166 -0)
📝 deep_gemm/__init__.py (+1 -0)
➕ deep_gemm/include/deep_gemm/impls/sm100_fp8_mqa_logits_f16_weights.cuh (+466 -0)
📝 tests/test_attention.py (+10 -2)

📄 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/DeepGEMM/pull/307 **Author:** [@benzh-2025](https://github.com/benzh-2025) **Created:** 4/17/2026 **Status:** ❌ Closed **Base:** `nv_dev` ← **Head:** `benzh/2cta_with_half` --- ### 📝 Commits (2) - [`f2d6a64`](https://github.com/deepseek-ai/DeepGEMM/commit/f2d6a6491d57ab1cd5bdd2bde3dec4a730d5afc0) add sm100_fp8_mqa_logits_f16_weights - [`6ec4c1d`](https://github.com/deepseek-ai/DeepGEMM/commit/6ec4c1d6f8cbe1f79408e920b7eb165c872fffe6) change test_mqa_logits to avoid f16 overflow when sm100_fp8_mqa_logits_f16_weights ### 📊 Changes **7 files changed** (+652 additions, -3 deletions) <details> <summary>View changed files</summary> 📝 `csrc/apis/attention.hpp` (+7 -1) 📝 `csrc/indexing/main.cu` (+1 -0) 📝 `csrc/jit_kernels/impls/runtime_utils.hpp` (+1 -0) 📝 `csrc/jit_kernels/impls/smxx_fp8_fp4_mqa_logits.hpp` (+166 -0) 📝 `deep_gemm/__init__.py` (+1 -0) ➕ `deep_gemm/include/deep_gemm/impls/sm100_fp8_mqa_logits_f16_weights.cuh` (+466 -0) 📝 `tests/test_attention.py` (+10 -2) </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

2026-10-26 21:16:36 +00:00