Skip to content

[PyTorch] Reduce NVFP4 quantization work for the fused GroupedMLP fc2 input with activation recompute - #3837

Open
xrennvidia wants to merge 5 commits into
NVIDIA:mainfrom
xrennvidia:xren/fused_grouped_mlp_rowwise_fc2x_recompute
Open

[PyTorch] Reduce NVFP4 quantization work for the fused GroupedMLP fc2 input with activation recompute#3837
xrennvidia wants to merge 5 commits into
NVIDIA:mainfrom
xrennvidia:xren/fused_grouped_mlp_rowwise_fc2x_recompute

Commits

  1. Commits on Nov 9, 2026