Feat/support nemotron h mtp#28541

Draft

shaharmor98 wants to merge 37 commits intovllm-project:mainfrom

shaharmor98:feat/support-nemotron-h-mtp

Commits on Nov 24, 2025

bugifx
shaharmor98
committed

Commits on Nov 25, 2025

working mxfp4
Tomer Natan
committed
config example
Tomer Natan
committed
block_scaled_matmul now expects prequantized A and B. It can handle both swizzled and un-swizzled scales
tomeras91
committed

Commits on Nov 27, 2025

Merge branch 'fake-mxfp8-moe' of ssh://gitlab-master.nvidia.com:12051/tasida/vllm into fake-mxfp8
Tomer Natan
committed
triton support for linear mxfp8
Tomer Natan
committed

Commits on Nov 30, 2025

add working real checkpoint code
shaharmor98
committed

Commits on Dec 1, 2025

Commits on Dec 2, 2025

BS1 prefix-caching works
roikoren755
committed
fix bug with mtp head selection
shaharmor98
committed
start working on triton
Tomer Natan
committed
Merge branch 'feat/support-nemotron-h-mtp' of github.com:shaharmor98/smor-vllm into linear-mxfp8-triton-support
Tomer Natan
committed
All prefix-caching works
roikoren755
committed
stable wip
Tomer Natan
committed
Merge pull request #2 from shaharmor98/feat/support-nemotron-h-mtp-prefix-caching
roikoren755
authored
Merge branch 'feat/support-nemotron-h-mtp' of github.com:shaharmor98/smor-vllm into linear-mxfp8-triton-support
Tomer Natan
committed
revert dequant
Tomer Natan
committed
remove quant from moe
Tomer Natan
committed

Commits on Dec 3, 2025

fix tp bug
Tomer Natan
committed

Commits on Dec 4, 2025

Merge pull request #6 from TomerBN-Nvidia/linear-mxfp8-triton-support
roikoren755
authored

Commits on Dec 8, 2025

support tp8
root
committed
pre commits
root
committed
Merge pull request #7 from TomerBN-Nvidia/support-tp-8-for-mxfp8
roikoren755
authored