Skip to content

[README] Update cuffpa-py library News🔥 #217

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Jan 10, 2025
Merged

[README] Update cuffpa-py library News🔥 #217

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
3dda091
Update README.md
DefTruth Jan 10, 2025
a8716e2
Update README.md
DefTruth Jan 10, 2025
b8d9baa
Update README.md
DefTruth Jan 10, 2025
6a51680
Update README.md
DefTruth Jan 10, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,8 @@
## 📖 News 🔥🔥
<div id="news"></div>

- [2025-01-08]: [📚Fully QKV Fine-grained Tiling](#mma-tiling-qkv) has been refactored into 🤖[cuffpa-py](https://github.com/DefTruth/cuffpa-py): 📚FFPA - Yet another Faster Flash Prefill Attention with O(1)🎉SRAM complexity for headdim > 256, ~1.5x🎉faster vs SDPA EA.
- [2025-01-08]: [📚QKV Fine-grained Tiling](#mma-tiling-qkv) has been refactored into 🤖[cuffpa-py](https://github.com/DefTruth/cuffpa-py): 📚FFPA - Yet another Faster Flash Prefill Attention with O(1)🎉SRAM complexity for headdim > 256, **1.5x~2x**🎉faster than SDPA EA: [📈L20 ~1.7x↑🎉](https://github.com/DefTruth/cuffpa-py?tab=readme-ov-file#L1-bench), [📈 A30 ~1.5x↑🎉](https://github.com/DefTruth/cuffpa-py?tab=readme-ov-file#L1-bench), [📈3080 ~2.5x↑🎉](https://github.com/DefTruth/cuffpa-py?tab=readme-ov-file#L1-bench), [📈4090 ~1.8x↑🎉](https://github.com/DefTruth/cuffpa-py?tab=readme-ov-file#L1-bench).

- [2024-12-02]: HGEMM MMA kernels has been refactored into 🤖[cuhgemm-py](https://github.com/DefTruth/cuhgemm-py): ⚡️Write HGEMM from scratch using Tensor Cores with WMMA, MMA and CuTe API, achieve peak⚡️ performance.

<!--
Expand All @@ -44,7 +45,7 @@
</div>


Currently, on NVIDIA L20, RTX 4090 and RTX 3080 Laptop, compared with cuBLAS's default Tensor Cores algorithm, the `HGEMM (WMMA/MMA/CuTe)` in this repo (`blue`🔵) can achieve `98%~100%` of its (`orange`🟠) performance. Please check [toy-hgemm library⚡️⚡️](./kernels/hgemm) or [hgemm-tensorcores-mma⚡️⚡️](https://github.com/DefTruth/hgemm-tensorcores-mma) repo for more details.
Currently, on NVIDIA L20, RTX 4090 and RTX 3080 Laptop, compared with cuBLAS's default Tensor Cores algorithm, the `HGEMM (WMMA/MMA/CuTe)` in this repo (`blue`🔵) can achieve `98%~100%` of its (`orange`🟠) performance. Please check [toy-hgemm library⚡️⚡️](./kernels/hgemm) or [cuhgemm-py⚡️⚡️](https://github.com/DefTruth/cuhgemm-py) repo for more details.

![toy-hgemm-library](https://github.com/user-attachments/assets/962bda14-b494-4423-b8eb-775da9f5503d)

Expand Down