Skip to content

[PyTorch] Grouped MLP: emit FP32 scale gradient with no cast to BF16 - #3870

Open
sraman-rgb wants to merge 11 commits into
NVIDIA:mainfrom
sraman-rgb:sraman/grouped-mlp-fp32-scales-grad
Open

sraman-rgb wants to merge 11 commits into
NVIDIA:mainfrom
sraman-rgb:sraman/grouped-mlp-fp32-scales-grad