tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1

Llama3 Swallow - Built with Meta Llama 3

Our Swallow model has undergone continual pre-training from the Llama 3 family, primarily with the addition of Japanese language data. The Instruct versions use supervised fine-tuning (SFT) and Chat Vector. Links to other models can be found in the index.

Model Release Updates

We are excited to share the release schedule for our latest models:

• July 1, 2024: Released the Llama-3-Swallow-8B-v0.1, Llama-3-Swallow-8B-Instruct-v0.1, Llama-3-Swallow-70B-v0.1, and Llama-3-Swallow-70B-Instruct-v0.1.

Swallow Model Index

Model	Llama-3-Swallow	Llama3 Swallow Instruct
8B	Link	Link
70B	Link	Link

This repository provides large language models developed by Swallow-LLM. Read our blog post.

Model Details

• Model type: Please refer to Llama 3 MODEL_CARD for details on the model architecture.
• Language(s): Japanese English
• Library: Megatron-LM
• Tokenizer: Please refer to Llama 3 blog for details on the tokenizer.
• Contact: swallow[at]nlp.c.titech.ac.jp

Model Performance

Japanese tasks

Model	Size	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
calm2-7b-chat	7B	0.2413	0.5128	0.4956	0.7729	0.0551	0.0480	0.2208	0.1384	0.2482	0.0000	0.2733
Swallow-7b-instruct-v0.1	7B	0.6059	0.4760	0.5284	0.8396	0.1546	0.1360	0.2285	0.1783	0.3510	0.0256	0.3524
Swallow-MS-7b-instruct-v0.1	7B	0.7435	0.5066	0.4268	0.8594	0.1582	0.1760	0.2260	0.1880	0.4177	0.2244	0.3927
RakutenAI-7B-chat	7B	0.9035	0.2600	0.4619	0.8647	0.1339	0.2120	0.2667	0.1966	0.4504	0.2299	0.3980
Qwen2-7B-Instruct	7B	0.8856	0.3902	0.3859	0.8967	0.1277	0.5720	0.2041	0.1909	0.5713	0.5683	0.4793
Meta-Llama-3-8B-Instruct	8B	0.8785	0.3812	0.3936	0.8955	0.1273	0.4160	0.2143	0.2035	0.4719	0.2872	0.4269
Llama-3-ELYZA-JP-8B	8B	0.9017	0.5124	0.5016	0.9113	0.1677	0.4600	0.2509	0.1846	0.4829	0.3811	0.4754
Llama-3-Swallow-8B-Instruct-v0.1	8B	0.9178	0.4963	0.5168	0.9088	0.1296	0.4880	0.2522	0.2254	0.4835	0.3927	0.4811

English tasks

Model	Size	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
calm2-7b-chat	7B	0.2860	0.3528	0.5042	0.2524	0.8413	0.3860	0.0546	0.2990	0.0000	0.3307
Swallow-7b-instruct-v0.1	7B	0.3280	0.4810	0.5501	0.2720	0.8774	0.4066	0.1251	0.3646	0.0866	0.3879
Swallow-MS-7b-instruct-v0.1	7B	0.3600	0.4999	0.5858	0.3030	0.8834	0.5273	0.2108	0.4386	0.2512	0.4511
RakutenAI-7B-chat	7B	0.4160	0.5971	0.6465	0.3091	0.8886	0.5757	0.3139	0.4958	0.2671	0.5011
Qwen2-7B-Instruct	7B	0.4000	0.5468	0.6146	0.3518	0.8852	0.7073	0.6300	0.3101	0.6354	0.5646
Meta-Llama-3-8B-Instruct	8B	0.3880	0.6687	0.5834	0.3743	0.8903	0.6567	0.7453	0.6478	0.5415	0.6107
Llama-3-ELYZA-JP-8B	8B	0.3200	0.5502	0.5224	0.3631	0.8809	0.5875	0.5701	0.3213	0.4604	0.5084
Llama-3-Swallow-8B-Instruct-v0.1	8B	0.3720	0.6557	0.5861	0.3648	0.9002	0.6315	0.5959	0.6391	0.4238	0.5743

MT-Bench JA

Model	Size	coding	extraction	humanities	math	reasoning	roleplay	stem	writing	JMTAvg
calm2-7b-chat	7B	0.1198	0.3793	0.4231	0.1011	0.1799	0.4760	0.3568	0.4583	0.3118
Swallow-7b-instruct-v0.1	7B	0.1947	0.3156	0.4991	0.1900	0.2141	0.5330	0.4535	0.4624	0.3578
Swallow-MS-7b-instruct-v0.1	7B	0.2235	0.3743	0.4611	0.1060	0.3404	0.4287	0.3969	0.3877	0.3398
RakutenAI-7B-chat	7B	0.2475	0.3522	0.4692	0.2140	0.3926	0.4427	0.3977	0.4434	0.3699
Qwen2-7B-Instruct	7B	0.4635	0.6909	0.6857	0.5970	0.5042	0.6667	0.5353	0.6808	0.6030
Meta-Llama-3-8B-Instruct	8B	0.3744	0.6876	0.6225	0.2070	0.5032	0.5248	0.5326	0.4884	0.4926
Llama-3-ELYZA-JP-8B	8B	0.2908	0.6421	0.6406	0.3088	0.5500	0.6740	0.5251	0.6744	0.5382
Llama-3-Swallow-8B-Instruct-v0.1	8B	0.3547	0.6508	0.5371	0.2718	0.4007	0.5493	0.4752	0.5730	0.4766

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])

MT-Bench JA

We used Japanese MT-Bench to assess the instruction-following capabilities of models. We utilized the following settings:

• Implemantation: FastChat [Zheng+, 2023] (commit #e86e70d0)
• Question: Nejumi LLM-Leaderboard NEO, mtbench_ja_question_v3
• Reference Answer: Nejumi LLM-Leaderboard NEO, mtbench_ja_referenceanswer_v1
• Prompt for Judge: Nejumi LLM-Lederboard NEO, mtbench_ja_prompt_v1
• Judge: gpt-4-1106-preview
• Scoring: Absolute scale normalized to a 0-1 range, averaged over five runs.

Usage

pip install vllm

from transformers import AutoTokenizer
from vllm import LLM, SamplingParams

model_name = "tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1"

tokenizer = AutoTokenizer.from_pretrained(model_name)
llm = LLM(
    model=model_name,
    tensor_parallel_size=1,
)

sampling_params = SamplingParams(
    temperature=0.6, top_p=0.9, max_tokens=512, stop="<|eot_id|>"
)


message = [
    {"role": "system", "content": "あなたは誠実で優秀な日本人のアシスタントです。"},
    {
        "role": "user",
        "content": "東京の夜空に打ち上がっている花火の下、向かい合っている燕とラマの温かい物語を書いてください。",
    },
]
prompt = tokenizer.apply_chat_template(
    message, tokenize=False, add_generation_prompt=True
)

output = llm.generate(prompt, sampling_params)

print(output[0].outputs[0].text)

Training Datasets

Instruction Tuning

The following datasets were used for the instruction tuning.

• OpenAssistant Conversations Dataset EN top-1 thread
• OpenAssistant Conversations Dataset was used, where human utterances are included but the responses are not used. Instead, the responses were generated using the Mixtral-8x7B-Instruct-v0.1 model.

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 under an open license for others to build on.

Our project is supported by the Large Generative AI Development Support Program of the National Institute of Advanced Industrial Science and Technology.

License

META LLAMA 3 COMMUNITY LICENSE

Authors

Here are the team members:

• From Tokyo Institute of Technology Okazaki Laboratory, the following members:
  • Naoaki Okazaki
  • Sakae Mizuki
  • Youmi Ma
  • Koki Maeda
  • Kakeru Hattori
  • Masanari Ohi
  • Taihei Shiotani
  • Koshiro Saito
• From Tokyo Institute of Technology YOKOTA Laboratory, the following members:
  • Rio Yokota
  • Kazuki Fujii
  • Taishi Nakamura
  • Takumi Okamoto
  • Ishida Shigeki
• From Artificial Intelligence Research Center, AIST, Japan, the following members:
  • Hiroya Takamura

How to cite

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

@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},
}

Citations

@article{llama3modelcard,
    title={Llama 3 Model Card},
    author={AI@Meta},
    year={2024},
    url = {https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md}
}