[MoE Refactor][16/N] Apply Refactor to NVFP4#31692

Merged

robertgshaw2-redhat merged 76 commits intomainfrom

Jan 8, 2026

Commits on Jan 4, 2026

initial commit
Robert Shaw
committed

Commits on Jan 5, 2026

refactoring to use the oracle
Robert Shaw
committed
nit
Robert Shaw
committed
raise error
Robert Shaw
committed
move not implemented error to __init__ from apply
Robert Shaw
committed
minor rename
Robert Shaw
committed
first working via flashinfer cutlass
Robert Shaw
committed
update
Robert Shaw
committed
update
Robert Shaw
committed
actually use apply
Robert Shaw
committed
we have correctness for flashinfer cutlass via the .apply() pathway
Robert Shaw
committed
convert to using the PFNoEP
Robert Shaw
committed
create the kernel in the oracle
Robert Shaw
committed
forgot to make the kernel :)
Robert Shaw
committed
last nit?
Robert Shaw
committed
minor comment change
Robert Shaw
committed
minor tweak to oracle
Robert Shaw
committed
remove flashinfer cutedsl from aply since it does not work via apply (it only does batched)
Robert Shaw
committed
flashinfer trtllm working properly again
Robert Shaw
committed
we are able to run with vllm cutlass, but have 0% accuracy score
Robert Shaw
committed
got accuracy with vllm cutlass
Robert Shaw
committed
convert to int comparison
Robert Shaw
committed
update comment
Robert Shaw
committed
update to decouple use_global_sf from ModelOpt
Robert Shaw
committed
initial attempt to apply this to compressed-tensors
Robert Shaw
committed
flashinfer cutlass working end-to-end with compressed-tensors
Robert Shaw
committed
updated
Robert Shaw
committed
fixed nvfp4 trtllm
Robert Shaw
committed
stash marlin fixes
Robert Shaw
committed
stash
Robert Shaw
committed
revert changes to ct
Robert Shaw
committed
first evidence of new marlin structure working
Robert Shaw
committed
marlin now working again for modelopt
Robert Shaw
committed
stash
Robert Shaw
committed
revert change for failing test
Robert Shaw
committed
test is now passing
Robert Shaw
committed
updated the --enforce-eager
Robert Shaw
committed
updated
Robert Shaw
committed
revert debug issues
Robert Shaw
committed
fixed up block_quant
Robert Shaw
committed
update comment
Robert Shaw
committed
stash changes
Robert Shaw
committed
update
Robert Shaw
committed
updated comments
Robert Shaw
committed
updated to remove debug nits
Robert Shaw
committed
updated
Robert Shaw
committed
Merge branch 'nvfp3-refactor' of https://github.com/vllm-project/vllm into nvfp4-refactor
Robert Shaw
committed
marlin nvfp4 working via modular kenrels for modelopt
Robert Shaw
committed
clean up
Robert Shaw
committed
replaced warning
Robert Shaw
committed
stash changes
Robert Shaw
committed
merge the bugfix
Robert Shaw
committed
compressed tensors nvfp4
Robert Shaw
committed
update comment
Robert Shaw
committed
fix the linter
Robert Shaw
committed
updated compressed-tensors
Robert Shaw
committed
fix dp/ep regression
Robert Shaw
committed
get past the linter
Robert Shaw
committed

Commits on Jan 6, 2026

Commits on Jan 7, 2026

updated
Robert Shaw
committed
update uvicorn acess
Robert Shaw
committed
update invocation of convert to kernel logic
Robert Shaw
committed
running
Robert Shaw
committed
remove cutlass_moe_fp4
Robert Shaw
committed
nit loc changes
Robert Shaw
committed
reduce loc changes
Robert Shaw
committed
fix missing import
Robert Shaw
committed
updated with minor nits to reduce LOC or eliminate debugs
Robert Shaw
committed

Commits on Jan 8, 2026

do the merge
Robert Shaw
committed
Update vllm/model_executor/layers/fused_moe/oracle/nvfp4.py

robertgshaw2-redhat
and
mgoin
authored
updated
Robert Shaw
committed
updated
Robert Shaw
committed
clean up --enforce-eager
Robert Shaw
committed
Update vllm/model_executor/layers/fused_moe/oracle/nvfp4.py

robertgshaw2-redhat
and
pavanimajety
authored