From d39b1d2adea1f1558c6eb0c692719e608b834e97 Mon Sep 17 00:00:00 2001 From: Anerudhan Gopal Date: Fri, 8 May 2026 13:02:51 -0700 Subject: [PATCH] Add acknowledgements section to sparse_attention.md Added acknowledgements for the Native Sparse Attention fprop kernels implementation. --- python/cudnn/native_sparse_attention/sparse_attention.md | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/python/cudnn/native_sparse_attention/sparse_attention.md b/python/cudnn/native_sparse_attention/sparse_attention.md index 8c57e6a4..95ddd005 100644 --- a/python/cudnn/native_sparse_attention/sparse_attention.md +++ b/python/cudnn/native_sparse_attention/sparse_attention.md @@ -158,3 +158,9 @@ Compression Attention and SWA support both T,H,D and B,H,S,D input formats. - **O (Output)**: `(B, H_q, S_q, D_v)` - **L (LogSumExp)**: `(B, H_q, S_q)` - **M (Max)**: `(B, H_q, S_q)` + +## Acknowledgements +The Native Sparse Attention fprop kernels, featuring support for head dim = 128 and optimized for the Blackwell architecture, was implemented using CuteDSL. + +These kernels was a collaborative effort jointly developed by: +Jie Feng, Akash Mehra, Vincent Zhang, Dominik Ernst, Xinbo Zhao, Aditya Vavre, Vedaanta Agarwalla, Mingyang Wang, Anerudhan Gopal, Paul Springer, Yang Xu, and Nima Tajbakhsh.