mlx-community/MiLMMT-46-12B-v0.1-4bit

This model mlx-community/MiLMMT-46-12B-v0.1-4bit was converted to MLX format from xiaomi-research/MiLMMT-46-12B-v0.1 using mlx-lm version 0.30.4.

Model Description

MiLMMT-46-12B-v0.1 is an LLM-based translation model by Xiaomi, finetuned from a continual pretrain of Gemma3-12B on 143 billion tokens of monolingual and parallel data across 46 languages. See the paper: Scaling Model and Data for Multilingual Machine Translation with Open Large Language Models.

• Supported Languages: Arabic, Azerbaijani, Bulgarian, Bengali, Catalan, Czech, Danish, German, Greek, English, Spanish, Persian, Finnish, French, Hebrew, Hindi, Croatian, Hungarian, Indonesian, Italian, Japanese, Kazakh, Khmer, Korean, Lao, Malay, Burmese, Norwegian, Dutch, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Swedish, Tamil, Thai, Tagalog, Turkish, Urdu, Uzbek, Vietnamese, Cantonese, Chinese (Simplified), Chinese (Traditional).
• GitHub: xiaomi-research/gemmax
• Developed by: Xiaomi Inc.

Translation Prompt

The model has no chat template. Use a raw completion prompt with English language names, tokenized without special tokens:

Translate this from <source language name> to <target language name>:
<source language name>: <source language sentence>
<target language name>:

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("mlx-community/MiLMMT-46-12B-v0.1-4bit")

text = "Translate this from Chinese (Simplified) to English:\nChinese (Simplified): 我爱机器翻译\nEnglish:"
prompt = tokenizer.encode(text, add_special_tokens=False)

response = generate(model, tokenizer, prompt=prompt, verbose=True)

License

Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms. This model is a derivative of Gemma 3 and is distributed under the same terms.

Citation

@misc{shang2026scalingmodeldatamultilingual,
      title={Scaling Model and Data for Multilingual Machine Translation with Open Large Language Models},
      author={Yuzhe Shang and Pengzhi Gao and Wei Liu and Jian Luan and Jinsong Su},
      year={2026},
      eprint={2602.11961},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2602.11961},
}