transformers
NVFP4: load a checkpoint's FP8 linears as FP8
#49922
Open

NVFP4: load a checkpoint's FP8 linears as FP8 #49922

IlyasMoutawwakil wants to merge 4 commits into nvfp4-experts from nvfp4-fp8-linears
IlyasMoutawwakil
IlyasMoutawwakil NVFP4: load FP8 linears as they ship (draft)
231c131d
IlyasMoutawwakil NVFP4: pass the FP8 linear's input as it is
853f4743
HuggingFaceDocBuilderDev
IlyasMoutawwakil NVFP4: start the expert stacks' serialized names at a name boundary
b47b2432
github-actions
IlyasMoutawwakil NVFP4: keep the checkpoint readers where they were
1bc7ce0e

Login to write a write a comment.

Login via GitHub

Reviewers
No reviews
Assignees
No one assigned
Labels
Milestone