Skip to content
CuFlash-Attn
OpenSpec-driven CUDA FlashAttention reference implementation with FP32/FP16 support, forward and backward kernels, and a stable v0.3.0 baseline aimed at long-term handoff and archival quality.
EnglishGuide, build notes, API reference, troubleshooting, and project status.简体中文面向中文读者的上手指南、构建说明、接口文档、故障排除与项目状态。
What this site is for
- explain the supported API and build surface
- point to the canonical OpenSpec design and verification sources
- provide a clean entry point for integration, review, and handoff work
Canonical project links
| Resource | Link |
|---|---|
| Repository | LessUp/cuflash-attn |
| Releases | GitHub Releases |
| OpenSpec specs | openspec/specs/ |