# thu-ml/SageAttention — RepoVerse

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

URL: https://repoverse.app/cards/867007699

[Repository source](https://github.com/thu-ml/SageAttention)

## Card snapshot

- Edition: Genesis
- Rarity: common
- Finish: Base
- Attack: 4732
- Defense: 3701
- GitHub stars: 3972
- GitHub forks: 534
- Language: Cuda
- Category: AI Pioneers
- Snapshot date: 2026-10-22

These repository statistics and game attributes belong to this dated card snapshot. They are not a current assessment of software quality.

This page describes the repository card, without a holder or individual copy record.

- [Rules](https://repoverse.app/rules.md)
- [Explore](https://repoverse.app/explore)