deepseek-ai / deepseek-ai/DeepGEMM
Will DeepGEMM support GEMM + 1d1d quantization all in one fused kernel?
Open
- Dominant language
- Cuda
- Stars
- 7.8k
- Forks
- 1.3k
- Avg merge
- 3d 7h
- Merged PRs (30d)
- 3
Description
Many thanks for the great work!
I was wondering if DeepGEMM will support GEMM + optional epilogue (e.g. elementwise or norm) + 1d1d post quantization all in one fused kernel? This should greatly improve the performance by eliminating the input activation quantization kernel for the next GEMM OP.
Kind regards
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.