| Demo for this model on Spaces | multimodalart | closed | 0 | disc |
| Update README | MaartenGr | merged | 0 | disc |
| Raise per-image vision soft-token budget from 280 to 1120 | lucianommartins | open | 0 | disc |
| Gemma 4 12B Unified (encoder-free audio): audio comprehension collapses past ~450–500 tokens of absolute position — vision on the same checkpoint is unaffected, E2B/E4B (Conformer) are unaffected | frandagostino | open | 0 | disc |
| SAE for this model? | joggyback | open | 0 | disc |
| Add response_template to tokenizer_config.json | Rocketknight1 | merged | 0 | disc |
| https://share.gemini.google/7T1rHMOMMfPH | DIGITALLEADERSHIP | open | 0 | disc |
| gemma-4-12B-it: deterministic "thought\n thought\n …" degenerate loop on long agent prompts (~60% reproducer, 4-bit, repros across temperatures) | Raullen | open | 0 | disc |
| Fairly Useless | yano2mch | open | 0 | disc |
| Gemma 4 12b bias affraid me | Alekto | open | 0 | disc |
| Chat template may re-inject prior-turn reasoning during multi-turn tool use → repetition loops | ManniX-ITA | open | 0 | disc |
| fdsfsdfsdfdf | jiwang1 | open | 0 | disc |
| asfd | jiwang1 | open | 0 | disc |
| fix: chat template — null handling, reasoning preservation, turn-tag balance, input validation | lucianommartins | merged | 0 | disc |
| Gemma 4 124b, please! | slavczays | open | 0 | disc |
| prompt template | maymo | closed | 0 | disc |
| vllm support? | mohamedemam | open | 0 | disc |
| gemma 4 12b falsesly claims year | Tralalabs | open | 0 | disc |
| Gemma4-12B Audio Deep Evaluation | PeterchanCN | open | 0 | disc |
| How to use an assistent model for MTP in llama.cpp? | Regrin | open | 0 | disc |
| Tool Calling and Tags and Context awareness. | JanjanJean | open | 0 | disc |
| MLX 4-bit conversion of gemma-4-12B-it — with vision + audio embedder weights preserved | jokernifty | open | 0 | disc |
| Gemma 4 124B DENSE + Nvidia N1x Laptop | 87455a | open | 0 | disc |
| What languages are supported in audio input? | J22 | open | 0 | disc |
| it is speaking Cantonese while user prompt is Simplified Chinese | leiyang88 | open | 0 | disc |
| Can't deploy on huggingface... | p-christ | open | 0 | disc |
| NVFP4 Quant | berkerdooo | open | 0 | disc |
| Update max_position_embeddings to full context length | dougreid | merged | 0 | disc |
| 👋 Training support for transformers/megatron backends | study-hjt | open | 0 | disc |
| NVFP4 / FP8 Quantizations | vincentzed-hf | open | 0 | disc |
| Dimension mismatch crashes the engine at startup | CallShaul | open | 0 | disc |
| Gemma 4 124b | FusionCow | open | 0 | disc |
| Thank you Google team | jreoka | open | 0 | disc |
| Thanks again! | JanjanJean | open | 0 | disc |
| MTP Assistant Model | patrickdeanbrown | open | 0 | disc |
| I've been having a bunch of issues trying to do translations from Chinese to English | unlambda | closed | 0 | disc |
| Preserve the correct ordering of content and tool calls when rendering OpenAI Chat Completions | yzong-rh | open | 0 | disc |
| fix max_position_embeddings 128Ki --> 256Ki | ddh0 | closed | 0 | disc |
| Wrong context size in config.json? | coder543 | closed | 0 | disc |
| Good, Google | dof-studio-org | open | 0 | disc |
| File path upper case issue | sleepyeldrazi | closed | 0 | disc |
| Add HLE evaluation result | SaylorTwift | open | 0 | disc |
| Add AIME 2026 evaluation result | SaylorTwift | open | 0 | disc |
| Add MMMU Pro evaluation result | SaylorTwift | open | 0 | disc |
| Add MMLU-Pro evaluation result | SaylorTwift | open | 0 | disc |
| Add GPQA Diamond evaluation result | SaylorTwift | open | 0 | disc |
| This is an awesome model proposition/idea for users with few VRAM. | nesymerp1 | open | 0 | disc |
| Gemma 4:124b | seamon67 | open | 0 | disc |