doesn’t support RTX 5080 (sm_120)
As soon as PyTorch adds support for sm_120, we can remove--use-cpu all --no-halfand enable full GPU acceleration. Do you know when support for sm 120 will be done?
The latest stable PyTorch 2.7.0 as well as the nightly binaries built with CUDA 12.8 support Blackwell architectures (sm_100, sm_120).