# DeepGEMM-Ascend: Fast GEMM Kernels for Huawei Ascend NPUs This repository profile is provided by osrepos.com, an open source repository discovery platform. Source: osrepos.com Repository profile: https://osrepos.com/repo/deepseek-ai-deepgemm-ascend Generated for open source discovery and AI-assisted research. DeepGEMM-Ascend ports DeepSeek's DeepGEMM library to Huawei Ascend NPUs, with API-compatible kernels for BF16, FP8 and FP4 GEMM, MQA logits, and MegaMoE. It targets teams optimizing inference and training workloads on Ascend 950 hardware. GitHub: https://github.com/deepseek-ai/DeepGEMM-Ascend OSRepos URL: https://osrepos.com/repo/deepseek-ai-deepgemm-ascend ## Summary DeepGEMM-Ascend ports DeepSeek's DeepGEMM library to Huawei Ascend NPUs, with API-compatible kernels for BF16, FP8 and FP4 GEMM, MQA logits, and MegaMoE. It targets teams optimizing inference and training workloads on Ascend 950 hardware. ## Topics - c++ - library - performance - ascend-npu - gemm-kernels - mixture-of-experts ## Repository Information Last analyzed by OSRepos: Sat Nov 14 2026 00:02:25 GMT+0000 (Western European Standard Time) Detail views: 2 GitHub clicks: 0 ## Safety Notice OSRepos shares public repositories for knowledge and discovery only. Review source code, dependencies, licenses, and security implications before running or installing anything. ## Content ## Overview DeepGEMM-Ascend is DeepSeek's port of its DeepGEMM matrix multiplication kernel library to Huawei Ascend NPUs. It provides a compatible API for Ascend users and addresses the complexity of implementing efficient kernels around Ascend's matrix multiply-add primitives, layouts, alignment constraints, and low-level parameters. The library targets performance-critical model workloads, combining a lightweight abstraction with Ascend-specific techniques such as sparse data loading and coroutine-based pipelining. Its documented support centers on Ascend 950 devices and includes dense GEMM, MQA logits, and MoE operations. ## Key Features - API-compatible port of DeepSeek's DeepGEMM, intended to preserve a familiar development workflow. - BF16, FP8, FP4, and mixed FP8×FP4 GEMM kernels. - MQA logits kernels and scheduling metadata utilities. - Grouped GEMM and MegaMoE operations for expert-based models. - Ascend-focused kernel techniques, including sparse data loading and coroutine-based pipelining. - Utilities for NPU architecture selection, core count, deterministic algorithms, and scaling-factor layout conversion. - JIT kernel compilation with configurable caching and debugging options. ## Use Cases - Ascend platform engineers needing low-level GEMM kernels for model inference or training. - DeepSeek model teams running MoE workloads that need grouped GEMM or fused MegaMoE operations. - Teams implementing MQA scoring for the DeepSeek Lightning Indexer on Ascend. - Kernel developers seeking reference implementations of performance optimization techniques for Ascend hardware. ## Our Take This is a focused, performance-oriented Ascend port with a compatible DeepGEMM interface, but its short project history and hardware-specific setup make it an early-stage choice for Ascend teams rather than a portable general-purpose library. **Good fit if:** - Your workloads run on Huawei Ascend 950 and need the listed GEMM, MQA, or MoE kernels. - You already use DeepGEMM APIs or want Ascend-specific kernel implementations to study. - Your team can build and validate C++ extensions in the documented CANN and Torch NPU environment. **Look elsewhere if:** - You need kernels for NVIDIA GPUs or another platform, rather than Huawei Ascend. - You cannot provide the required Ascend hardware, CANN toolkit, and Torch NPU package. - You need a project with a longer release history, since this repository is new and has no recorded release in the past year. ## Project Health | Signal | Value | |---|---| | Status | **New**: less than six months old | | Last commit | 2026-10-14 (30 days ago) | | Commits, last 90 days | 3 | | Releases, last 12 months | 0 | | Contributors | 3 (top contributor: 33% of commits) | | Issues closed, last 90 days | 2 | | Pull requests merged, last 90 days | 1 | | Project age | 45 days | Checked on 2026-11-14 with the GitHub API. ## Project Facts - Language: C++ - License: MIT - Stars: 639 - Forks: 85 - Archived: no ## What You Need Detected in the repository: - Python (from setup.py) - A test suite ## Getting Started Clone the repository with its submodules, then install the package: ```bash git clone --recursive https://github.com/deepseek-ai/DeepGEMM-Ascend.git cd DeepGEMM-Ascend pip install . --no-build-isolation ``` See the [README](https://github.com/deepseek-ai/DeepGEMM-Ascend#readme) for requirements, development setup, interfaces, and benchmark details. ## License in Plain Words **MIT** (permissive). - Commercial use: yes - Modify and redistribute: yes - You must keep: the copyright and license notice - Share your changes: no A summary, not legal advice: the LICENSE file is what applies. ## Considerations - The documented development and validation target is Ascend 950, with CANN 9.20 and `torch_npu`; this is not a drop-in choice for other accelerator platforms. - Building from source requires Python 3.10 or newer, C++20 `` support, and the listed compiler tools. The HC prenorm kernel also depends on TileLang. - The project is very new. Recent activity is limited, with only a few commits in the measured 90-day period and no releases in the past year. Three contributors are recorded, with work not concentrated in one person. - The repository includes tests, but its measured project data shows no CI configuration. Teams should validate their own hardware and software combinations. - The MIT license permits broad use, subject to its license terms and attribution requirements.