LLaMAX2: Your Translation-Enhanced Model also Performs Well in Reasoning
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909834126819328 |
|---|---|
| author | Gao, Changjiang Huang, Zixian Gong, Jingyang Huang, Shujian Li, Lei Yuan, Fei |
| author_facet | Gao, Changjiang Huang, Zixian Gong, Jingyang Huang, Shujian Li, Lei Yuan, Fei |
| contents | General Large Language Models (LLMs) excel in reasoning, but those enhanced for translation struggle with reasoning tasks. To address this, we propose a novel translationenhanced recipe that begins with instruct models and applies layer-selective tuning only on parallel data. Following this pipeline, we introduce the Qwen3-XPlus models, which demonstrate significant improvements in translation performance across both high- and lowresource languages, achieving 15+ spBLEU and 40+ xComet in low-resource languages, like Swahili. Interestingly, training only with small parallel datasets, Qwen3-XPlus achieves an average improvement of 1+ points on 7 multilingual tasks while maintaining proficiency comparable to the Qwen3 instruct model in 15 popular reasoning datasets. This work offers a promising approach to multilingual enhancement, significantly reducing complexity and enhancing accessibility for a wider range of languages. The code and model are publicly available. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_09189 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | LLaMAX2: Your Translation-Enhanced Model also Performs Well in Reasoning Gao, Changjiang Huang, Zixian Gong, Jingyang Huang, Shujian Li, Lei Yuan, Fei Computation and Language General Large Language Models (LLMs) excel in reasoning, but those enhanced for translation struggle with reasoning tasks. To address this, we propose a novel translationenhanced recipe that begins with instruct models and applies layer-selective tuning only on parallel data. Following this pipeline, we introduce the Qwen3-XPlus models, which demonstrate significant improvements in translation performance across both high- and lowresource languages, achieving 15+ spBLEU and 40+ xComet in low-resource languages, like Swahili. Interestingly, training only with small parallel datasets, Qwen3-XPlus achieves an average improvement of 1+ points on 7 multilingual tasks while maintaining proficiency comparable to the Qwen3 instruct model in 15 popular reasoning datasets. This work offers a promising approach to multilingual enhancement, significantly reducing complexity and enhancing accessibility for a wider range of languages. The code and model are publicly available. |
| title | LLaMAX2: Your Translation-Enhanced Model also Performs Well in Reasoning |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2510.09189 |