[Cherry-pick] PRs #1801 #1808 #1629 #1627 #1824 #1826 #1830 #1760 #1831 #1858 #1839 #1857 #1869 - #1880

Merged

kevalmorabia97 merged 12 commits into

release/0.45.0from

cherry-picks/release-0.45.0

Jul 2, 2026

Commits on Jul 1, 2026

Fix real quant backend import cycle (#1801)

mxinO
authored and
kevalmorabia97
committed
[6281412] docs: update TensorRT-Edge-LLM CLI commands in torch_onnx example (#1808)

authored and
kevalmorabia97
committed
remove deprecated get_default_load_sharded_strategy (#1629)

dimapihtar
authored and
kevalmorabia97
committed
[6058907] Fix ShapeInferenceError in ONNX int8+fp16 quantization of weakly-typed models (#1627)

authored and
kevalmorabia97
committed
Fix reduce_amax NotImplementedError on FP8 weights (NVBug 6360175) (#1824)

authored and
kevalmorabia97
committed
fix(recipes): exclude Llama-4 vision branch from default PTQ quantization (#1826)

authored and
kevalmorabia97
committed
using validation keyword both in puzzletron configs and in dataset pr… (#1830)

grzegorz-k-karch
and
kevalmorabia97
committed
feat(recipes): add nvfp4_mlp_only-novit-kv_fp8 (exclude VL vision tower) (#1760)

authored and
kevalmorabia97
committed
Fix lm_eval_hf freezing issue on multi-gpu slurm interactive node (#1831)

danielkorzekwa
authored and
kevalmorabia97
committed
fix(export): list unquantized MoE routers in exclude_modules (NVBug 5718750) (#1858)

authored and
kevalmorabia97
committed
build(docs): pin sphinx-argparse<0.6.0 to fix doc build error

kevalmorabia97
and
claude
committed
Back-port HF PTQ dtype/init-config fixes (#1839, #1857, #1869)

kevalmorabia97
and
claude
committed