[Model] Add Gemma3 GGUF multimodal support#27772

Merged

vllm-bot merged 53 commits intovllm-project:mainfrom

lucianommartins:main

Nov 18, 2025

Commits on Nov 17, 2025

Add Gemma3 GGUF multimodal support
lucianommartins
committed
Remove deprecated V0 compatibility code
lucianommartins
committed
Address architectural feedback:
lucianommartins
committed
Address feedbacks
lucianommartins
committed
Fix ReadTheDocs type annotations in utils.py
lucianommartins
committed
Fix type annotation imports - move to module level
lucianommartins
committed
fix: restore model_arch parameter for GGUF dtype handling
lucianommartins
committed
fix: restore model_arch parameter for GGUF dtype handling
lucianommartins
committed
Address reviewer feedback: remove hardcoded values, apply fail-fast
lucianommartins
committed
refactor: eliminate code duplication and generalize multimodal GGUF support
lucianommartins
committed
refactor: address code review feedback for Gemma3 GGUF
lucianommartins
committed
Addressing reviews/feedbacks.
lucianommartins
committed
reverting cosmetic changes.
lucianommartins
committed
refactor: reorganize GGUF utilities and extract config patching
lucianommartins
committed
Add Gemma3 GGUF multimodal generation tests
lucianommartins
committed
test: split Gemma3 tests into separate GGUF and HF files
lucianommartins
committed
clean

Isotr0py
authored and
lucianommartins
committed
avoid reading GGUF multiple times

Isotr0py
authored and
lucianommartins
committed
better hf_config patch

Isotr0py
authored and
lucianommartins
committed
clean test

Isotr0py
authored and
lucianommartins
committed
remove redundant vibe coding test

Isotr0py
authored and
lucianommartins
committed
update autom tensor mapping

Isotr0py
authored and
lucianommartins
committed
gguf: implement automatic mmproj weight mapping with filtering
lucianommartins
committed
fix(gguf): resolve Gemma3 multimodal parameter prefix mismatch
lucianommartins
committed
fix(gguf): resolve Gemma3 multimodal parameter prefix mismatch
lucianommartins
committed
bump gguf version

Isotr0py
authored and
lucianommartins
committed
refactor(gguf): use official GGUF constants for vision config extraction
lucianommartins
committed
clean

Isotr0py
authored and
lucianommartins
committed
fix Gemma3 GGUF multimodal test with correct processor loading
lucianommartins
committed
unify unqunatized weights handling

Isotr0py
authored and
lucianommartins
committed
revert unnecessary changes

Isotr0py
authored and
lucianommartins
committed
update gemma3mm embed_input_ids

Isotr0py
authored and
lucianommartins
committed
fix

Isotr0py
authored and
lucianommartins
committed
clean model config

Isotr0py
authored and
lucianommartins
committed
clean processor to load processor from tokenizer

Isotr0py
authored and
lucianommartins
committed
feat(gguf): Finish gguf loader code cleanup
lucianommartins
committed
move tokenizer validation to model_config

Isotr0py
authored and
lucianommartins
committed
fix test

Isotr0py
authored and
lucianommartins
committed
revert gemma3 text backbone

Isotr0py
authored and
lucianommartins
committed
compatability with unsloth gguf

Isotr0py
authored and
lucianommartins
committed
fix broken inc quant

Isotr0py
authored and
lucianommartins
committed
revert unnecessary changes

Isotr0py
authored and
lucianommartins
committed
revert unnecessary changes

Isotr0py
authored and
lucianommartins
committed
clean and correct some comments

Isotr0py
authored and
lucianommartins
committed
update test

Isotr0py
authored and
lucianommartins
committed
fix deadlock

Isotr0py
authored and
lucianommartins
committed
gemini

Isotr0py
authored and
lucianommartins
committed
Add vocab_size override and improve automatic weight mapping
lucianommartins
committed
fix: Update CI test configurations for Gemma3 compatibility
lucianommartins
committed
refactor(gguf): remove redundant vocab_size extraction
lucianommartins
committed
fix(v1): add hasattr check before calling generate_attention_masks
lucianommartins
committed

Commits on Nov 18, 2025