thu-ml / SageAttention
[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.
- Repository stars
- 3,997All-time #15,415
- Fork
- 548
- Last push
- updated 9 months ago
- Repository created
- since 2024
About
Summarised by AI from the README; the original prevails
Author description: [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.
Basic listing based on official public information; functionality has not been independently evaluated.
Check before adopting
- No pushes for over 6 months (last 2026-01-17); check whether it is maintained.
- Platform
- Not verified
- Pricing
- Pricing not verifiedCheck current official software, hosting, model and infrastructure pricing.
- Maintenance notes
- Not verified