Acknowledgements

CuTe DSL OSS kernels

Several open-source kernels in this repository originated in NVIDIA's CuTe DSL kernel library. We gratefully acknowledge the contributors who helped develop, bring up, and test these kernels.

Contributors include:

• Rachit Garg rachitg@nvidia.com
• Jack Yang jackyang@nvidia.com
• Hao Sheng hsheng@nvidia.com
• Alex Li alel@nvidia.com
• Aragorn Guan aragorng@nvidia.com
• Bangyu Shen bangyus@nvidia.com
• Caleb Du cadu@nvidia.com
• Xiao Song xiaos@nvidia.com
• Siddhartha Raman sraman@nvidia.com

This acknowledgement covers the CuTe DSL kernel work now represented by these modules in python/cudnn/:

• gemm_amax
• gemm_dsrelu
• gemm_srelu
• gemm_swiglu
• grouped_gemm/grouped_gemm_dglu
• grouped_gemm/grouped_gemm_dsrelu
• grouped_gemm/grouped_gemm_dswiglu
• grouped_gemm/grouped_gemm_glu
• grouped_gemm/grouped_gemm_glu_hadamard
• grouped_gemm/grouped_gemm_quant
• grouped_gemm/grouped_gemm_srelu
• grouped_gemm/grouped_gemm_swiglu
• grouped_gemm/grouped_gemm_wgrad
• discrete_grouped_gemm/discrete_grouped_gemm_dswiglu
• discrete_grouped_gemm/discrete_grouped_gemm_swiglu
• rmsnorm_rht_amax

Thank you also to the broader CUTLASS/CuTe DSL and infrastructure teams who supported the original kernel development.