The LMT aims to advance the frontier of Multilingual Machine Translation (MMT) by building Inclusive, Scalable, and High-performance multilingual translation models.

• 2025.11.11: Our LMT paper Beyond English: Toward Inclusive and Scalable Multilingual Machine Translation with LLMs and corresponding Models are released.

We have made the following resources available:

In this project, we take a step toward overcoming the prevailing English-centric bias in MMT. We introduce LMT, a suite of Chinese-English-centric MMT models trained on 90B mixed monolingual and bilingual tokens, covering 60 languages across 234 translation directions and achieving SOTA performance among models with similar language coverage. Our work makes the following main contributions:

• We identify and analyze a previously overlooked issue, directional degeneration, in large-scale multilingual SFT with multi-way data and propose a simple yet effective Strategic Downsampling method to mitigate it.
• We propose Parallel Multilingual Prompting (PMP), which enhances cross-lingual transfer by incorporating an auxiliary parallel sentence into the instruction.
• We release LMT, a suite of large-scale Chinese–English-centric multilingual translation models in four sizes (0.6B/1.7B/4B/8B), providing strong baselines for future MMT research.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "NiuTrans/LMT-60-8B"

tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side='left')
model = AutoModelForCausalLM.from_pretrained(model_name)

prompt = "Translate the following text from English into Chinese.
English: The concept came from China where plum blossoms were the flower of choice.
Chinese: "
messages = [{"role": "user", "content": prompt}]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(**model_inputs, max_new_tokens=512, num_beams=5, do_sample=False)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

outputs = tokenizer.batch_decode(output_ids, skip_special_tokens=True)

print("response:", outputs)

For more details, please refer to src/inference.py.

Email: [email protected]

If you find our paper useful for your research, please kindly cite our paper:

@misc{luoyf2025lmt,
      title={Beyond English: Toward Inclusive and Scalable Multilingual Machine Translation with LLMs}, 
      author={Yingfeng Luo, Ziqiang Xu, Yuxuan Ouyang, Murun Yang, Dingyang Lin, Kaiyan Chang, Tong Zheng, Bei Li, Peinan Feng, Quan Du, Tong Xiao, Jingbo Zhu},
      year={2025},
      eprint={2511.07003},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2511.07003}, 
}