Skip to content

[PyTorch Debug] Add scale_inv_std stat and skip NVFP4 layers in LogFp8TensorStats - #3044

Closed
pggPL wants to merge 12 commits into
NVIDIA:mainfrom
pggPL:debug_log_fp_fixes
Closed

pggPL wants to merge 12 commits into
NVIDIA:mainfrom
pggPL:debug_log_fp_fixes