---
license: gemma
language:
  - mi
  - en
base_model: google/gemma-3-12b-pt
tags:
  - byol
  - low-resource
  - māori
  - gemma3
library_name: transformers
pipeline_tag: text-generation
---

# BYOL Māori 12B

This model was produced by the [BYOL framework](https://github.com/microsoft/byol)
for extending LLMs to low-resource languages.

- **Base model:** [google/gemma-3-12b-pt](https://huggingface.co/google/gemma-3-12b-pt)
- **Language:** Māori (mri)
- **Training stage:** Merged (CPT + IT via model merging)
- **License:** [Gemma Terms of Use](https://ai.google.dev/gemma/terms) (derived from Gemma 3)
- **Paper:** [BYOL: Bring Your Own Language Into LLMs](https://arxiv.org/abs/2601.10804)
- **Code:** [github.com/microsoft/byol](https://github.com/microsoft/byol)

## Model Description

This is a **merged** language model for Māori (mri) that combines the language knowledge acquired during continual pre-training with the instruction-following capabilities from supervised fine-tuning. It was produced by merging [BYOL Māori 12b CPT](https://huggingface.co/ai-for-good-lab/byol-mri-12b-cpt) and [BYOL Māori 12b IT](https://huggingface.co/ai-for-good-lab/byol-mri-12b-it) checkpoints back into the original Gemma 3 instruction model, using the [BYOL framework](https://github.com/microsoft/byol).

**This is the recommended model for most users.** It supports chat/instruction-following and has the strongest overall performance on Māori benchmarks (see the [paper](https://arxiv.org/abs/2601.10804) for evaluation results).

## Usage

```bash
pip install -U transformers
```

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "ai-for-good-lab/byol-mri-12b-merged"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", dtype=torch.bfloat16)

# Chat inference
messages = [{"role": "user", "content": "Kōrerotia mai mō Aotearoa."}]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True, return_dict=True).to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Citation

```bibtex
@article{zamir2026byolbringlanguagellms,
    title={BYOL: Bring Your Own Language Into LLMs},
    author={Syed Waqas Zamir and Wassim Hamidouche and Boulbaba Ben Amor and Luana Marotti and Inbal Becker-Reshef and Juan Lavista Ferres},
    year={2026},
    journal={arXiv:2601.10804},
    url={https://arxiv.org/abs/2601.10804},
}
```