[PyTorch] Preserve fprop operands for dequantized backward override#3141

ptrendx merged 4 commits into

NVIDIA:mainNVIDIA/TransformerEngine:mainfrom

negvet:fix_dequantized_override_save_original_inputnegvet/TransformerEngine:fix_dequantized_override_save_original_inputCopy head branch name to clipboard