Building a Large Japanese Web Corpus for Large Language Models
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_ | 1866913333202911232 |
|---|---|
| author | Okazaki, Naoaki Hattori, Kakeru Shota, Hirai Iida, Hiroki Ohi, Masanari Fujii, Kazuki Nakamura, Taishi Loem, Mengsay Yokota, Rio Mizuki, Sakae |
| author_facet | Okazaki, Naoaki Hattori, Kakeru Shota, Hirai Iida, Hiroki Ohi, Masanari Fujii, Kazuki Nakamura, Taishi Loem, Mengsay Yokota, Rio Mizuki, Sakae |
| contents | Open Japanese large language models (LLMs) have been trained on the Japanese portions of corpora such as CC-100, mC4, and OSCAR. However, these corpora were not created for the quality of Japanese texts. This study builds a large Japanese web corpus by extracting and refining text from the Common Crawl archive (21 snapshots of approximately 63.4 billion pages crawled between 2020 and 2023). This corpus consists of approximately 312.1 billion characters (approximately 173 million pages), which is the largest of all available training corpora for Japanese LLMs, surpassing CC-100 (approximately 25.8 billion characters), mC4 (approximately 239.7 billion characters) and OSCAR 23.10 (approximately 74 billion characters). To confirm the quality of the corpus, we performed continual pre-training on Llama 2 7B, 13B, 70B, Mistral 7B v0.1, and Mixtral 8x7B Instruct as base LLMs and gained consistent (6.6-8.1 points) improvements on Japanese benchmark datasets. We also demonstrate that the improvement on Llama 2 13B brought from the presented corpus was the largest among those from other existing corpora. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2404_17733 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Building a Large Japanese Web Corpus for Large Language Models Okazaki, Naoaki Hattori, Kakeru Shota, Hirai Iida, Hiroki Ohi, Masanari Fujii, Kazuki Nakamura, Taishi Loem, Mengsay Yokota, Rio Mizuki, Sakae Computation and Language Artificial Intelligence Open Japanese large language models (LLMs) have been trained on the Japanese portions of corpora such as CC-100, mC4, and OSCAR. However, these corpora were not created for the quality of Japanese texts. This study builds a large Japanese web corpus by extracting and refining text from the Common Crawl archive (21 snapshots of approximately 63.4 billion pages crawled between 2020 and 2023). This corpus consists of approximately 312.1 billion characters (approximately 173 million pages), which is the largest of all available training corpora for Japanese LLMs, surpassing CC-100 (approximately 25.8 billion characters), mC4 (approximately 239.7 billion characters) and OSCAR 23.10 (approximately 74 billion characters). To confirm the quality of the corpus, we performed continual pre-training on Llama 2 7B, 13B, 70B, Mistral 7B v0.1, and Mixtral 8x7B Instruct as base LLMs and gained consistent (6.6-8.1 points) improvements on Japanese benchmark datasets. We also demonstrate that the improvement on Llama 2 13B brought from the presented corpus was the largest among those from other existing corpora. |
| title | Building a Large Japanese Web Corpus for Large Language Models |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2404.17733 |