Multilingual Pretraining Using a Large Corpus Machine-Translated from a Single Source Language
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912106734944256 |
|---|---|
| author | Wang, Jiayi Lu, Yao Weber, Maurice Ryabinin, Max Chen, Yihong Tang, Raphael Stenetorp, Pontus |
| author_facet | Wang, Jiayi Lu, Yao Weber, Maurice Ryabinin, Max Chen, Yihong Tang, Raphael Stenetorp, Pontus |
| contents | English, as a very high-resource language, enables the pretraining of high-quality large language models (LLMs). The same cannot be said for most other languages, as leading LLMs still underperform for non-English languages, likely due to a gap in the quality and diversity of the available multilingual pretraining corpora. In this work, we find that machine-translated text from a single high-quality source language can contribute significantly to the pretraining of multilingual LLMs. We translate FineWeb-Edu, a high-quality English web dataset, into French, German, and Spanish, resulting in a final 300B-token dataset, which we call TransWeb-Edu, and pretrain a 1.3B-parameter model, CuatroLLM, from scratch on this dataset. Across five non-English reasoning tasks, we show that CuatroLLM matches or outperforms state-of-the-art multilingual models trained using closed data, such as Llama3.2 and Gemma2, despite using an order of magnitude less data, such as about 6% of the tokens used for Llama3.2's training. We further demonstrate that with additional domain-specific pretraining, amounting to less than 1% of TransWeb-Edu, CuatroLLM surpasses the state of the art in multilingual reasoning. To promote reproducibility, we release our corpus, models, and training pipeline under open licenses at hf.co/britllm/CuatroLLM. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_23956 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Multilingual Pretraining Using a Large Corpus Machine-Translated from a Single Source Language Wang, Jiayi Lu, Yao Weber, Maurice Ryabinin, Max Chen, Yihong Tang, Raphael Stenetorp, Pontus Computation and Language English, as a very high-resource language, enables the pretraining of high-quality large language models (LLMs). The same cannot be said for most other languages, as leading LLMs still underperform for non-English languages, likely due to a gap in the quality and diversity of the available multilingual pretraining corpora. In this work, we find that machine-translated text from a single high-quality source language can contribute significantly to the pretraining of multilingual LLMs. We translate FineWeb-Edu, a high-quality English web dataset, into French, German, and Spanish, resulting in a final 300B-token dataset, which we call TransWeb-Edu, and pretrain a 1.3B-parameter model, CuatroLLM, from scratch on this dataset. Across five non-English reasoning tasks, we show that CuatroLLM matches or outperforms state-of-the-art multilingual models trained using closed data, such as Llama3.2 and Gemma2, despite using an order of magnitude less data, such as about 6% of the tokens used for Llama3.2's training. We further demonstrate that with additional domain-specific pretraining, amounting to less than 1% of TransWeb-Edu, CuatroLLM surpasses the state of the art in multilingual reasoning. To promote reproducibility, we release our corpus, models, and training pipeline under open licenses at hf.co/britllm/CuatroLLM. |
| title | Multilingual Pretraining Using a Large Corpus Machine-Translated from a Single Source Language |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2410.23956 |