Hugging Face's logo

tokyotech-llm
/
Llama-3.1-Swallow-8B-v0.2

Text Generation
Transformers
Safetensors
English
Japanese
llama
text-generation-inference
Inference Endpoints
Model card Files Community

Llama 3.1 Swallow - Built with Llama

Llama 3.1 Swallow is a series of large language models (8B, 70B) that were built by continual pre-training on the Meta Llama 3.1 models. Llama 3.1 Swallow enhanced the Japanese language capabilities of the original Llama 3.1 while retaining the English language capabilities. We use approximately 200 billion tokens that were sampled from a large Japanese web corpus (Swallow Corpus Version 2), Japanese and English Wikipedia articles, and mathematical and coding contents, etc for continual pre-training. The instruction-tuned models (Instruct) were built by supervised fine-tuning (SFT) on the synthetic data specially built for Japanese (see the Training Datasets section for details). See the Swallow Model Index section to find other model variants.

Release History

Swallow Model Index

Model Llama-3.1-Swallow v0.1 Llama-3.1-Swallow-Instruct v0.1 Llama-3.1-Swallow v0.2 Llama-3.1-Swallow-Instruct v0.2 Llama-3.1-Swallow-Instruct v0.3
8B Link Link Link Link Link
70B Link Link

logo

The website https://swallow-llm.github.io/ provides large language models developed by the Swallow team.

Model Details

  • Model type: Please refer to Llama 3.1 MODEL_CARD for details on the model architecture.
  • Language(s): Japanese English
  • Library: Megatron-LM
  • Tokenizer: Please refer to Llama 3.1 blog for details on the tokenizer.
  • Contact: swallow[at]nlp.c.titech.ac.jp

Model Performance

Japanese tasks

Model JCom. JEMHopQA NIILC JSQuAD XL-Sum MGSM WMT20-en-ja WMT20-ja-en JMMLU JHumanEval Ja Avg
4-shot 4-shot 4-shot 4-shot 1-shot 4-shot 4-shot 4-shot 5-shot 0-shot
EM acc Char-F1 Char-F1 Char-F1 ROUGE-2 EM acc BLEU BLEU EM acc pass@1
Qwen2-7B 0.8776 0.4627 0.3766 0.8984 0.1716 0.5480 0.2080 0.1949 0.5871 0.4183 0.4743
Qwen2.5-7B 0.9240 0.4581 0.4259 0.9071 0.2162 0.6200 0.2295 0.1989 0.6337 0.2665 0.4880
Sarashina2-7B 0.7417 0.5089 0.6353 0.8683 0.1420 0.0800 0.2727 0.2015 0.3835 0.0000 0.3834
Llama 3 8B 0.8356 0.4454 0.4002 0.8881 0.1757 0.3320 0.2199 0.2087 0.4558 0.3311 0.4292
Llama 3.1 8B 0.8436 0.4461 0.4050 0.8962 0.1794 0.3560 0.2209 0.2077 0.4767 0.3274 0.4359
Llama 3 Youko 8B 0.8660 0.4902 0.5155 0.8947 0.2127 0.2840 0.2740 0.2180 0.4493 0.2183 0.4423
Llama 3 Swallow 8B 0.8945 0.4848 0.5640 0.8947 0.1981 0.4240 0.2758 0.2223 0.4699 0.2890 0.4717
Llama 3.1 Swallow 8B v0.1 0.9124 0.5092 0.6011 0.8991 0.2020 0.4600 0.2909 0.2313 0.5182 0.2811 0.4905
Llama 3.1 Swallow 8B v0.2 0.9106 0.5097 0.6272 0.8922 0.1976 0.4640 0.2957 0.2326 0.5253 0.3360 0.4991

English tasks

Model OpenBookQA TriviaQA HellaSWAG SQuAD2.0 XWINO MMLU GSM8K BBH HumanEval En Avg
4-shot 4-shot 4-shot 4-shot 4-shot 5-shot 4-shot 3-shot 0-shot
Acc EM acc Acc EM acc Acc Acc EM acc CoT EM Acc pass@1
Qwen2-7B 0.3740 0.6105 0.6006 0.3623 0.8916 0.7045 0.7748 0.5325 0.4622 0.5903
Qwen2.5-7B 0.3940 0.6011 0.5999 0.3743 0.8890 0.7424 0.8324 0.5620 0.4213 0.6018
Sarashina2-7B 0.3420 0.4784 0.5327 0.2911 0.8903 0.4267 0.1008 0.3746 0.0000 0.3818
Llama 3 8B 0.3760 0.7109 0.6124 0.3356 0.9032 0.6509 0.4936 0.6211 0.3793 0.5648
Llama 3.1 8B 0.3780 0.7017 0.6094 0.3330 0.9045 0.6525 0.5057 0.6176 0.3695 0.5636
Llama 3 Youko 8B 0.3500 0.6252 0.5885 0.3247 0.8959 0.5993 0.3571 0.5704 0.2793 0.5100
Llama 3 Swallow 8B 0.3520 0.6563 0.5901 0.3507 0.9006 0.6152 0.4875 0.5936 0.3323 0.5420
Llama 3.1 Swallow 8B v0.1 0.3800 0.6711 0.6057 0.3468 0.9032 0.6237 0.5110 0.6153 0.3622 0.5577
Llama 3.1 Swallow 8B v0.2 0.3820 0.6510 0.5955 0.3473 0.9041 0.6227 0.5208 0.6053 0.3659 0.5549

Evaluation Benchmarks

Japanese evaluation benchmarks

We used llm-jp-eval(v1.3.0), JP Language Model Evaluation Harness(commit #9b42d41) and Code Generation LM Evaluation Harness(commit #0261c52). The details are as follows:

  • Multiple-choice question answering (JCommonsenseQA [Kurihara et al., 2022])
  • Open-ended question answering (JEMHopQA [Ishii et al., 2024])
  • Open-ended question answering (NIILC [関根, 2003])
  • Machine reading comprehension (JSQuAD [Kurihara et al., 2022])
  • Automatic summarization (XL-Sum [Hasan et al., 2021])
  • Machine translation (WMT2020 ja-en [Barrault et al., 2020])
  • Machine translation (WMT2020 en-ja [Barrault et al., 2020])
  • Mathematical reasoning (MGSM [Shi et al., 2023])
  • Academic exams (JMMLU [尹ら, 2024])
  • Code generation (JHumanEval [佐藤ら, 2024])

English evaluation benchmarks

We used the Language Model Evaluation Harness(v.0.4.2) and Code Generation LM Evaluation Harness(commit #0261c52). The details are as follows:

  • Multiple-choice question answering (OpenBookQA [Mihaylov et al., 2018])
  • Open-ended question answering (TriviaQA [Joshi et al., 2017])
  • Machine reading comprehension (SQuAD2 [Rajpurkar et al., 2018])
  • Commonsense reasoning (XWINO [Tikhonov and Ryabinin, 2021])
  • Natural language inference (HellaSwag [Zellers et al., 2019])
  • Mathematical reasoning (GSM8K [Cobbe et al., 2021])
  • Reasoning (BBH (BIG-Bench-Hard) [Suzgun et al., 2023])
  • Academic exams (MMLU [Hendrycks et al., 2021])
  • Code generation (HumanEval [Chen et al., 2021])

Training Datasets

Continual Pre-Training

The following datasets were used for continual pre-training.

Swallow Corpus Version 2

We built the Swallow Corpus by extracting high-quality Japanese texts from Common Crawl. In Version 2, we expanded the scope of the Common Crawl collection and modified the pipeline sequence to enable more flexible quality filtering. For Llama 3.1 Swallow v0.2, we further refined our quality filtering and data sampling strategies, resulting in an even higher-quality selection of Japanese texts for pre-training.

Further details of the methodology and analysis will be provided in a forthcoming paper.

The-stack-v2(filtered)

We created a high-quality Python code corpus, The-Stack-v2(filtered), by applying filtering to The-Stack-v2-train-smol-ids. This filtering process utilizes Python's compile() function, pylint, and language detection on comments within the code to select only data that meets a certain quality threshold. Further details will be available in our forthcoming paper.

Risks and Limitations

The models released here are still in the early stages of our research and development and have not been tuned to ensure outputs align with human intent and safety considerations.

Acknowledgements

We thank Meta Research for releasing Llama 3.1 under a generous open license.

We received various supports including:

  • AIST project: "Research and Development of Foundation Models for Generative AI in the Physical Domain"
  • NEDO project: "Development of Artificial Intelligence Application Technology to Support Judgment in Design Risk Assessment Work Based on the Perspective of Skilled Persons" (JPNP18002) of "Development of Integration Technology as the Core of Next Generation Artificial Intelligence and Robotics"
  • MEXT project: "Formation of R&D center to ensure transparency and reliability of generative AI models"
  • AIST program: Large Generative AI Development Support Program

License

META LLAMA 3.1 COMMUNITY LICENSE

Authors

Here are the team members:

How to cite

If you find our work helpful, please feel free to cite these papers.

@inproceedings{Fujii:COLM2024,
   title={Continual Pre-Training for Cross-Lingual LLM Adaptation:
Enhancing Japanese Language Capabilities},
   author={Kazuki Fujii and Taishi Nakamura and Mengsay Loem and Hiroki
Iida and Masanari Ohi and Kakeru Hattori and Hirai Shota and Sakae
Mizuki and Rio Yokota and Naoaki Okazaki},
   booktitle="Proceedings of the First Conference on Language Modeling",
   series={COLM},
   pages="(to appear)",
   year="2024",
   month=oct,
   address={University of Pennsylvania, USA},
}

@inproceedings{Okazaki:COLM2024,
   title={Building a Large Japanese Web Corpus for Large Language Models},
   author={Naoaki Okazaki and Kakeru Hattori and Hirai Shota and Hiroki
Iida and Masanari Ohi and Kazuki Fujii and Taishi Nakamura and Mengsay
Loem and Rio Yokota and Sakae Mizuki},
   booktitle="Proceedings of the First Conference on Language Modeling",
   series={COLM},
   pages="(to appear)",
   year="2024",
   month=oct,
   address={University of Pennsylvania, USA},
}

References 

@misc{dubey2024llama3herdmodels,
      title={The Llama 3 Herd of Models}, 
      author={Abhimanyu Dubey and Abhinav Jauhri and Abhinav Pandey and Abhishek Kadian and Ahmad Al-Dahle and Aiesha Letman and Akhil Mathur and Alan Schelten and Amy Yang and Angela Fan et al.},
      year={2024},
      eprint={2407.21783},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2407.21783}, 
}
Downloads last month
1,807
Safetensors
Model size
8.03B params
Tensor type
BF16
·
Inference Examples
Text Generation
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for tokyotech-llm/Llama-3.1-Swallow-8B-v0.2

Merges
1 model
Quantizations
3 models

Collection including tokyotech-llm/Llama-3.1-Swallow-8B-v0.2