Gemma 3 / Gemma 3n

Gemma 3 includes text-only and multimodal checkpoints. Gemma 3n is a next-generation efficiency-focused variant. This is the single coverage page for the Gemma 3 generation.


Task	Text Generation, Image-Text-to-Text
Architecture	`Gemma3ForCausalLM`, `Gemma3ForConditionalGeneration`
Parameters	270M – 27B
HF Org	google

Available Models

Gemma 3 270M (text-only)
Gemma 3 27B IT (VL)
Gemma 3 4B IT (VL)
Gemma 3n 4B (VL)

Architecture

Gemma3ForCausalLM — text-only models
Gemma3ForConditionalGeneration — multimodal models

Example HF Models

Model	HF ID
Gemma 3 270M	`google/gemma-3-270m`
Gemma 3 4B IT	`google/gemma-3-4b-it`
Gemma 3 27B IT	`google/gemma-3-27b-it`

Example Recipes

Recipe	Dataset	Description
gemma_3_270m_squad.yaml	SQuAD	SFT — Gemma 3 270M
gemma_3_270m_squad_peft.yaml	SQuAD	LoRA — Gemma 3 270M
gemma3_vl_4b_cord_v2.yaml	cord-v2	SFT — Gemma 3 4B VL on CORD-v2
gemma3_vl_4b_cord_v2_peft.yaml	cord-v2	LoRA — Gemma 3 4B VL on CORD-v2
gemma3_vl_4b_cord_v2_megatron_fsdp.yaml	cord-v2	SFT — Gemma 3 4B VL with MegatronFSDP
gemma3_vl_4b_medpix.yaml	MedPix-VQA	SFT — Gemma 3 4B VL on MedPix
gemma3n_vl_4b_medpix.yaml	MedPix-VQA	SFT — Gemma 3n 4B VL on MedPix
gemma3n_vl_4b_medpix_peft.yaml	MedPix-VQA	LoRA — Gemma 3n 4B VL on MedPix

Try with NeMo AutoModel

1. Clone and install from source (full instructions):

$ git clone https://github.com/NVIDIA-NeMo/Automodel.git
$ cd Automodel
$ uv sync --locked --all-groups --all-extras

2. Run the recipe from inside the repo:

$ uv run automodel --nproc-per-node=8 examples/vlm_finetune/gemma3/gemma3_vl_4b_cord_v2.yaml

Run with Docker

1. Pull the container and mount a checkpoint directory:

$ docker run --gpus all -it --rm \
>   --shm-size=8g \
>   -v $(pwd)/checkpoints:/opt/Automodel/checkpoints \
>   nvcr.io/nvidia/nemo-automodel:26.06.00

2. Navigate to the AutoModel directory (where the recipes are):

$ cd /opt/Automodel

3. Run the recipe:

$ automodel --nproc-per-node=8 examples/vlm_finetune/gemma3/gemma3_vl_4b_cord_v2.yaml

See the Installation Guide and VLM Fine-Tuning Guide.

Fine-Tuning

See the Gemma 3 & Gemma 3n Fine-Tuning Guide for detailed instructions on dataset preparation, configuration, and multi-GPU training.