[PyTorch] Support for cuDNN-backed flex attention#2984

vcherepanov-nv wants to merge 16 commits into

NVIDIA:mainNVIDIA/TransformerEngine:mainfrom

vcherepanov-nv:cudnn-score-mod-3vcherepanov-nv/TransformerEngine:cudnn-score-mod-3Copy head branch name to clipboard