Update Cutlass to 4.0

Author: tridao

.github/workflows/publish.yml

@@ -144,7 +144,7 @@ jobs:
           export LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
           # Limit MAX_JOBS otherwise the github runner goes OOM
           # nvcc 11.8 can compile with 2 jobs, but nvcc 12.3 goes OOM
-          MAX_JOBS=$([ "$MATRIX_CUDA_VERSION" == "128" ] && echo 1 || echo 2) FLASH_ATTENTION_FORCE_BUILD="TRUE" FLASH_ATTENTION_FORCE_CXX11_ABI=${{ matrix.cxx11_abi}} python setup.py bdist_wheel --dist-dir=dist
+          MAX_JOBS=$([ "$MATRIX_CUDA_VERSION" == "129" ] && echo 1 || echo 2) FLASH_ATTENTION_FORCE_BUILD="TRUE" FLASH_ATTENTION_FORCE_CXX11_ABI=${{ matrix.cxx11_abi}} python setup.py bdist_wheel --dist-dir=dist
           tmpname=cu${WHEEL_CUDA_VERSION}torch${MATRIX_TORCH_VERSION}cxx11abi${{ matrix.cxx11_abi }}
           wheel_name=$(ls dist/*whl | xargs -n 1 basename | sed "s/-/+$tmpname-/2")
           ls dist/*whl |xargs -I {} mv {} dist/${wheel_name}