MXFP8 on the fine-grained machinery (`Mxfp8Config`) #49780
IlyasMoutawwakil
wants to merge
14 commits into
fg-split/7-nvfp4
from
fg-split/8-mxfp8
Add Mxfp8Config: MXFP8 on the fine-grained machinery
3511ee66
Document MXFP8, MiniMax-M3's public checkpoints and the formats compr…
8e2afa28
Patch the finish helpers where the quantizer imports them
e08eb979
Align the overview table row
a46e96bd
Point the MiniMax-M3-VL speed example at the bf16 repo
dcdb6351
Build the MiniMax-M3-VL parallel fixture from its model tester
806fbc6b
Write the comments and docstrings in short plain sentences
ae606dc2
Drop W8A4 from the MXFP8 docs and mixed-scheme grid, and document its…
1d10679d
Pin MiniMax-M3-VL's routing in the parallel fixture and restore its d…
cabe18c9
Explain why the FP8 quantizer import stays at the top of auto.py
a714ffb6
Give the table test a model that holds the table
c4985f46
Read MiniMax's ignored_layers through an mxfp8 reader
be47b9fc
Leave MXFP8's ModelOpt and compressed-tensors layouts to the producer…
25ce4e30
Restack on the simpler plan resolution and merging
42b862c5
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub