TORCH_CUDA_ARCH_LIST="8.9;12.0" pip install --no-cache-dir --no-build-isolation flash-attn and FLASH_ATTENTION_BUILD_FROM_SOURCE=1 TORCH_CUDA_ARCH_LIST="8.9;12.0" pip ...
Sorry for my dumb question but how to install intel-extension-for-pytorch since from venv python 3.10/12 python -m pip install intel-extension-for-pytorch oneccl_bind ...