deepseek-ai/DeepGEMM
DeepGEMM: clean and efficient BLAS kernel library on GPU
Build with Backblaze B2
SDKs, agent skills, IDE extensions, and reference pipelines from Backblaze Labs. All open source.
Loading star history...
Where it stands
Rank, growth pace and release cadence relative to everything GStars tracks.
No releases in the last 90 days; the most recent was 106 days ago, with 45 contributors on GitHub.
Peaked at #36 in the weekly ranking the week of Apr 19, its best in the past year.
Use Cases & Benefits
- Provides a high-performance CUDA tensor core kernel library for GEMM operations optimized for NVIDIA GPUs supporting FP8, FP4, BF16 data types and fused MoE computations.
- Delivers expert-tuned performance with a lightweight, clean codebase and runtime JIT compilation, avoiding heavy template dependencies and enabling fast kernel loading without CUDA recompilation.
- Use for accelerating large language model training and inference by optimizing matrix multiplications with mixed precision on NVIDIA SM90/SM100 GPUs.
- Use for implementing fused Mixture of Experts (MoE) models with overlapped communication and computation to maximize throughput in multi-process distributed setups.
- Use for developing custom GPU kernels requiring flexible grouped GEMM layouts and advanced tensor core utilization with minimal CPU overhead and fast JIT compilation.
Should you depend on it?
License terms, project hygiene and risks, from GitHub metadata and GStars' snapshots.
License
MIT
Permissive: commercial use, modification and redistribution are allowed with attribution.
Community files
- README: present
Watch-outs
No watch-outs from GStars' checks.
General guidance from repository metadata, not legal advice. How we score
From the README
DeepGEMM is a unified, high-performance tensor core kernel library that brings together the key computation primitives of modern large language models — GEMMs (FP8, FP4, BF16), fused MoE with overlapped communication (Mega MoE), MQA scoring for the lightning indexer, HyperConnection (HC), and more — into a single, cohesive CUDA codebase. All kernels are compiled at runtime via a lightweight Just-In-Time (JIT) module, requiring no CUDA compilation during installation.
DeepGEMM leverages some concepts from CUTLASS and CuTe, but avoids heavy reliance on their templates or algebras. The library is designed for simplicity, with only a limited number of core kernel functions, making it a clean and accessible resource for learning NVIDIA GPU kernel optimization techniques.
Despite its lightweight design, DeepGEMM's performance matches or exceeds expert-tuned libraries across various matrix shapes.
Daily snapshots
Stars gained per day, and how forks and open issues moved, from GStars' daily snapshots
+1,511 stars over 86 days across 8 snapshot days; best day +144 on Oct 7.
Forks
1.5K+347
Open issues
169+63
Compared with peers in Cuda
Fastest-growing repositories GStars tracks in this cohort for the week of Nov 1, by 7-day star growth
| Repository | Stars | +7d | Health | Last release |
|---|---|---|---|---|
| deepseek-ai/DeepEP | 10.4K | — | 35 | 416 days ago |
| deepseek-ai/DeepGEMM | 9.2K | — | 46 | 106 days ago |
| flashinfer-ai/flashinfer | 6.8K | — | 60 | 1 day ago |
| MoonshotAI/FlashKDA | 1.3K | — | 8 | — |
| Peer median | 6.8K | — | 35 | — |
Also by deepseek-ai
Other repositories from this maintainer that GStars tracks
- Python104.7K stars
DeepSeek-V3
- C++13.3K stars
FlashMLA
FlashMLA: Efficient Multi-head Latent Attention Kernels
- Cuda10.4K stars
DeepEP
DeepEP: an efficient expert-parallel communication library
- 6.3K stars
awesome-deepseek-agent
- Python4.8K stars
Engram
Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models
Discover Repositories
Search across tracked repositories by name or description