[None][feat] Add Gemma 4 12B Unified (encoder-free multimodal) support - #15768

Merged

Hudayday merged 13 commits into

NVIDIA:mainfrom

Hudayday:feat/gemma4-12b-unified

Jul 4, 2026

Commits on Jun 30, 2026

[None][feat] Add Gemma 4 12B Unified (gemma4_unified) support
Hudayday
committed
8120447
[None][test] Fix Gemma4Unified unit test (mapper-registry lookup + __new__ attrs)
Hudayday
committed
f7de1d5
[None][test] Skip gemma4_unified test when transformers<5.10
Hudayday
committed
6e6cc9e
Merge branch 'main' of https://github.com/NVIDIA/TensorRT-LLM into feat/gemma4-12b-unified
Hudayday
committed
3317304
[None][chore] Apply pre-commit (isort + ruff-format) to gemma4_unified
Hudayday
committed
4452b4f
[None][fix] Gemma4: keep audio causal in mm bidirectional mask (match HF)
Hudayday
committed
ccb3070

Commits on Jul 1, 2026

[None][feat] gemma4-12b-unified: address PR review feedback
Hudayday
committed
3b16ff6

Commits on Jul 2, 2026

Commits on Jul 3, 2026

[None][fix] gemma4-12b-unified: back the inherited mm_token_ids property
Hudayday
committed
de5812f