# thu-ml/SageAttention

<https://repo.masters-masters.com/en/repos/thu-ml/SageAttention>

[Home](https://repo.masters-masters.com/en) › [thu-ml](https://repo.masters-masters.com/en/owners/thu-ml) › [SageAttention](https://repo.masters-masters.com/en/repos/thu-ml/SageAttention)

thu-ml/SageAttention has 3,991 stars on GitHub. Its RepoMasters card: RC tier. \[ICLR2025, ICML2025, NeurIPS2025 Spotlight\] Quantized Attention achieves…

| Field | Value |
| --- | --- |
| Owner | [thu-ml](https://repo.masters-masters.com/en/owners/thu-ml) |
| Stars | 3,991 |
| Forks | 546 |
| Language | [Cuda](https://repo.masters-masters.com/en/languages/cuda) |
| License | Apache-2.0 |
| Created | 2024 |
| Topics | attention, inference-acceleration, llm, quantization, cuda, triton, video-generation, efficient-attention, mlsys, llm-infra |
| Tier | [RC (2,000 to 4,999 ★)](https://repo.masters-masters.com/en/ranks/blockbuster) |
| View on GitHub | [thu-ml/SageAttention](https://github.com/thu-ml/SageAttention) |

## Repository

\[ICLR2025, ICML2025, NeurIPS2025 Spotlight\] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

The tier for repos with 2,000 to 4,999 ★. Published odds per card pulled: 3%. Rarity: Epic.

## More repos by thu-ml

- [thu-ml/tianshou · ★ 11,046 · Python · Stable](https://repo.masters-masters.com/en/repos/thu-ml/tianshou)
- [thu-ml/TurboDiffusion · ★ 3,857 · Python · RC](https://repo.masters-masters.com/en/repos/thu-ml/TurboDiffusion)
- [thu-ml/zhusuan · ★ 2,218 · Python · RC](https://repo.masters-masters.com/en/repos/thu-ml/zhusuan)
- [thu-ml/RoboticsDiffusionTransformer · ★ 1,816 · Python · Beta](https://repo.masters-masters.com/en/repos/thu-ml/RoboticsDiffusionTransformer)
- [thu-ml/prolificdreamer · ★ 1,563 · Python · Beta](https://repo.masters-masters.com/en/repos/thu-ml/prolificdreamer)
- [thu-ml/unidiffuser · ★ 1,489 · Python · Beta](https://repo.masters-masters.com/en/repos/thu-ml/unidiffuser)

## Nearby in Cuda

- [leoxiaobin/deep-high-resolution-net.pytorch · ★ 4,476 · Cuda · RC](https://repo.masters-masters.com/en/repos/leoxiaobin/deep-high-resolution-net.pytorch)
- [Infatoshi/cuda-course · ★ 4,168 · Cuda · RC](https://repo.masters-masters.com/en/repos/Infatoshi/cuda-course)
- [baidu-research/warp-ctc · ★ 4,065 · Cuda · RC](https://repo.masters-masters.com/en/repos/baidu-research/warp-ctc)
- [HazyResearch/ThunderKittens · ★ 3,792 · Cuda · RC](https://repo.masters-masters.com/en/repos/HazyResearch/ThunderKittens)
- [hujie-frank/SENet · ★ 3,653 · Cuda · RC](https://repo.masters-masters.com/en/repos/hujie-frank/SENet)
- [BBuf/how-to-optim-algorithm-in-cuda · ★ 3,340 · Cuda · RC](https://repo.masters-masters.com/en/repos/BBuf/how-to-optim-algorithm-in-cuda)

---
Data: GitHub · Illustration: the project’s public image, pixelated by RepoMasters.