LLaMAX: Scaling Linguistic Horizons of LLM by Enhancing Translation Capabilities Beyond 100 Languages
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910645848375296 |
|---|---|
| author | Lu, Yinquan Zhu, Wenhao Li, Lei Qiao, Yu Yuan, Fei |
| author_facet | Lu, Yinquan Zhu, Wenhao Li, Lei Qiao, Yu Yuan, Fei |
| contents | Large Language Models (LLMs) demonstrate remarkable translation capabilities in high-resource language tasks, yet their performance in low-resource languages is hindered by insufficient multilingual data during pre-training. To address this, we conduct extensive multilingual continual pre-training on the LLaMA series models, enabling translation support across more than 100 languages. Through a comprehensive analysis of training strategies, such as vocabulary expansion and data augmentation, we develop LLaMAX. Remarkably, without sacrificing its generalization ability, LLaMAX achieves significantly higher translation performance compared to existing open-source LLMs (by more than 10 spBLEU points) and performs on-par with specialized translation model (M2M-100-12B) on the Flores-101 benchmark. Extensive experiments indicate that LLaMAX can serve as a robust multilingual foundation model. The code \footnote{\url{https://github.com/CONE-MT/LLaMAX/.}} and the models \footnote{\url{https://huggingface.co/LLaMAX/.}} are publicly available. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2407_05975 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | LLaMAX: Scaling Linguistic Horizons of LLM by Enhancing Translation Capabilities Beyond 100 Languages Lu, Yinquan Zhu, Wenhao Li, Lei Qiao, Yu Yuan, Fei Computation and Language Artificial Intelligence Large Language Models (LLMs) demonstrate remarkable translation capabilities in high-resource language tasks, yet their performance in low-resource languages is hindered by insufficient multilingual data during pre-training. To address this, we conduct extensive multilingual continual pre-training on the LLaMA series models, enabling translation support across more than 100 languages. Through a comprehensive analysis of training strategies, such as vocabulary expansion and data augmentation, we develop LLaMAX. Remarkably, without sacrificing its generalization ability, LLaMAX achieves significantly higher translation performance compared to existing open-source LLMs (by more than 10 spBLEU points) and performs on-par with specialized translation model (M2M-100-12B) on the Flores-101 benchmark. Extensive experiments indicate that LLaMAX can serve as a robust multilingual foundation model. The code \footnote{\url{https://github.com/CONE-MT/LLaMAX/.}} and the models \footnote{\url{https://huggingface.co/LLaMAX/.}} are publicly available. |
| title | LLaMAX: Scaling Linguistic Horizons of LLM by Enhancing Translation Capabilities Beyond 100 Languages |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2407.05975 |