Library of CUTLASS kernels targeting Large Language Models (LLM).

1. Download CUTLASS following instructions from: https://github.com/NVIDIA/cutlass.
2. Modify the (hardcoded) path in the sample compile.sh to your CUTLASS directory.
3. Run the modified compile.sh as ./compile.sh.

1. While running the executable make sure to set NVIDIA_TF32_OVERRIDE=1 to enable TF32 mode for cuBLAS for SGEMM. Otherwise, cuBLAS uses float32.

1. See README.md in sub-directories for more specific instructions.