[PyTorch] Build the attention padding mask without host syncs - #3268

[PyTorch] Build the attention padding mask without host syncs#3268

pggPL merged 2 commits into

NVIDIA:mainNVIDIA/TransformerEngine:mainfrom

pggPL:padding_mask_no_host_syncpggPL/TransformerEngine:padding_mask_no_host_syncCopy head branch name to clipboard