Repository navigation

[OMNIML-5858] Add a Nemotron 3.5 Super VL AutoQuant recipe - #2582

Merged

meenchen merged 7 commits into

mainfrom

weimingc/autoquant-mtp-recipes

Oct 15, 2026

Merged

[OMNIML-5858] Add a Nemotron 3.5 Super VL AutoQuant recipe#2582
meenchen merged 7 commits into
mainfrom
weimingc/autoquant-mtp-recipes

Commits on Oct 15, 2026
- feat(autoquant): add fixed MTP mixed-precision recipes
  meenchen
  committed
  5facaa9
- refactor(recipe): keep one MSE AutoQuant recipe with fixed MTP
  meenchen
  committed
  341415f
- fix(recipe): cast KV and search standalone Nemotron-H decoders
  meenchen
  committed
  321360e
- fix(recipe): limit MTP quantization to routed experts
  meenchen
  committed
  1ea057c
- fix(recipe): calibrate MTP KV and quantize shared experts
  meenchen
  committed
  c916c0d
- refactor(recipe): scope AutoQuant to Nemotron 3.5 Super VL
  meenchen
  committed
  8ca7e5b
- docs(recipe): omit optional AutoQuant README
  meenchen
  committed
  d241cef