[PR #240] Add n shape bias #306

New issue

Open

opened 2026-10-26 21:16:04 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:16:04 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/240
Author: @xueweilnvidia
Created: 12/17/2025
Status: 🔄 Open

Base: nv_dev ← Head: add_n_shape_bias

📝 Commits (6)

cfe29c4 minor
d96986f accumulate and bias true works
6e38032 minor
7dd9fbc update smem
e314311 minor
3ae9b1a bug fix

📊 Changes

10 files changed (+203 additions, -47 deletions)

View changed files

📝 csrc/apis/attention.hpp (+1 -1)
📝 csrc/apis/gemm.hpp (+28 -8)
📝 csrc/jit_kernels/heuristics/common.hpp (+10 -3)
📝 csrc/jit_kernels/impls/runtime_utils.hpp (+19 -0)
📝 csrc/jit_kernels/impls/sm100_fp8_gemm_1d1d.hpp (+18 -7)
📝 deep_gemm/include/deep_gemm/impls/sm100_fp8_gemm_1d1d.cuh (+89 -7)
📝 tests/generators.py (+24 -9)
📝 tests/test_attention.py (+1 -1)
📝 tests/test_bf16.py (+5 -5)
📝 tests/test_fp8.py (+8 -6)

📄 Description

Add bias with shape [n] support to fp8 gemm on sm100. Only support bf16 output datatype

test with:
test_attention.py test_bf16.py test_fp8.py

the performance are in attached files
performance.txt

_{🔄 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/240 **Author:** [@xueweilnvidia](https://github.com/xueweilnvidia) **Created:** 12/17/2025 **Status:** 🔄 Open **Base:** `nv_dev` ← **Head:** `add_n_shape_bias` --- ### 📝 Commits (6) - [`cfe29c4`](https://github.com/deepseek-ai/DeepGEMM/commit/cfe29c47913ede8d1b9213063728489d35c882c6) minor - [`d96986f`](https://github.com/deepseek-ai/DeepGEMM/commit/d96986f627c0ddb0a2d315eb7816735308413555) accumulate and bias true works - [`6e38032`](https://github.com/deepseek-ai/DeepGEMM/commit/6e38032647e4fc8dc6c0440ae4bea46f6504ce19) minor - [`7dd9fbc`](https://github.com/deepseek-ai/DeepGEMM/commit/7dd9fbc719861dc5d8068e5813d2ca2a3b714d8f) update smem - [`e314311`](https://github.com/deepseek-ai/DeepGEMM/commit/e314311edbf7423cdce64e13d1a820e1b6124b6b) minor - [`3ae9b1a`](https://github.com/deepseek-ai/DeepGEMM/commit/3ae9b1a2fa4062cb73583250b7509bfae6d10e14) bug fix ### 📊 Changes **10 files changed** (+203 additions, -47 deletions) <details> <summary>View changed files</summary> 📝 `csrc/apis/attention.hpp` (+1 -1) 📝 `csrc/apis/gemm.hpp` (+28 -8) 📝 `csrc/jit_kernels/heuristics/common.hpp` (+10 -3) 📝 `csrc/jit_kernels/impls/runtime_utils.hpp` (+19 -0) 📝 `csrc/jit_kernels/impls/sm100_fp8_gemm_1d1d.hpp` (+18 -7) 📝 `deep_gemm/include/deep_gemm/impls/sm100_fp8_gemm_1d1d.cuh` (+89 -7) 📝 `tests/generators.py` (+24 -9) 📝 `tests/test_attention.py` (+1 -1) 📝 `tests/test_bf16.py` (+5 -5) 📝 `tests/test_fp8.py` (+8 -6) </details> ### 📄 Description Add bias with shape [n] support to fp8 gemm on sm100. Only support bf16 output datatype test with: test_attention.py test_bf16.py test_fp8.py the performance are in attached files [performance.txt](https://github.com/user-attachments/files/24726642/performance.txt) --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi added the

pull-request

label

2026-10-26 21:16:04 +00:00

No labels

pull-request

No milestone

No project

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/DeepGEMM#306

No description provided.

Rows
Columns