[Cherry-pick] PRs #1256 #1305 #1322 #1317 #1321 #1289 #1311 #1332 #1104 #1318 - #1350

Merged

kevalmorabia97 merged 10 commits into

release/0.44.0from

cherry-picks/release-0.44.0

Apr 27, 2026

Commits on Apr 27, 2026

Exclude small-k and small-n Matmul nodes from Int8 quantization (#1256)

nv-samcheng
authored and
kevalmorabia97
committed
Fix vLLM fakequant MoE megatron export bug (#1305)

kinjalpatel27
authored and
kevalmorabia97
committed
docs: update installation pages with legal-approved license notices (#1322)

kevalmorabia97
and
claude
committed
Update vLLM deployment docs for heterogeneous models (#1317)

grzegorz-k-karch
and
kevalmorabia97
committed
Fix lm_eval version checking (#1321)
kevalmorabia97
committed
[OMNIML-3349] Add FP8 MHA quantization support for HuggingFace ViT (#1289)

ajrasane
authored and
kevalmorabia97
committed
fix: bug hf_ptq.py max_length setting ignored for LLMs (#1311)

michaelfeil
authored and
kevalmorabia97
committed
Deprecate Mllama support in llm_ptq/vlm_ptq examples (#1332)

cjluo-nv
authored and
kevalmorabia97
committed
fix: EAGLE mix_hidden_states in-place op crash (#1088) (#1104)

javierdejesusda
authored and
kevalmorabia97
committed
Fix PTQ for VLMs with image calibration (#1318)

LianaMikael
authored and
kevalmorabia97
committed