richarderkhov/tokyotech-llm_-_llama-3-swallow-70b-instruct-v0.1-gguf overview

{
  "metadata": {},
  "card_data": {
    "frontmatter": {},
    "hero_image_url": "./logo.png",
    "summary": "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:",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "Quantization made by Richard Erkhov.\n\n[Github](https://github.com/RichardErkhov)\n\n[Discord](https://discord.gg/pvy7H8DZMG)\n\n[Request more models](https://github.com/RichardErkhov/quant_request)\n\n\nLlama-3-Swallow-70B-Instruct-v0.1 - GGUF\n- Model creator: https://huggingface.co/tokyotech-llm/\n- Original model: https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-Instruct-v0.1/\n\n\n| Name | Quant method | Size |\n| ---- | ---- | ---- |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q2_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.Q2_K.gguf) | Q2_K | 24.56GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.IQ3_XS.gguf) | IQ3_XS | 27.29GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.IQ3_S.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.IQ3_S.gguf) | IQ3_S | 28.79GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.Q3_K_S.gguf) | Q3_K_S | 28.79GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.IQ3_M.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.IQ3_M.gguf) | IQ3_M | 29.74GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q3_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.Q3_K.gguf) | Q3_K | 31.91GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.Q3_K_M.gguf) | Q3_K_M | 31.91GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.Q3_K_L.gguf) | Q3_K_L | 34.59GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.IQ4_XS.gguf) | IQ4_XS | 35.64GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q4_0.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/blob/main/Llama-3-Swallow-70B-Instruct-v0.1.Q4_0.gguf) | Q4_0 | 37.22GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | IQ4_NL | 37.58GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q4_K_S | 37.58GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q4_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q4_K | 39.6GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q4_K_M | 39.6GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q4_1.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q4_1 | 41.27GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q5_0.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q5_0 | 45.32GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q5_K_S | 45.32GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q5_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q5_K | 46.52GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q5_K_M | 46.52GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q5_1.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q5_1 | 49.36GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q6_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q6_K | 53.91GB |\n| [Llama-3-Swallow-70B-Instruct-v0.1.Q8_0.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf/tree/main/) | Q8_0 | 69.83GB |\n\n\n\n\nOriginal model description:\n---\nlanguage:\n  - en\n  - ja\nlibrary_name: transformers\npipeline_tag: text-generation\nlicense: llama3\nmodel_type: llama\n---\n\n# Llama3 Swallow\n\nOur Swallow model has undergone continual pre-training from the [Llama 3 family](https://huggingface.co/collections/meta-llama/meta-llama-3-66214712577ca38149ebb2b6), 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.\n\n\n# Model Release Updates\n\nWe are excited to share the release schedule for our latest models:\n- **July 1, 2024**: Released the [Llama-3-Swallow-8B-v0.1](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-8B-v0.1), [Llama-3-Swallow-8B-Instruct-v0.1](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1), [Llama-3-Swallow-70B-v0.1](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-v0.1), and [Llama-3-Swallow-70B-Instruct-v0.1](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-Instruct-v0.1).\n\n## Swallow Model Index\n\n|Model|Llama-3-Swallow|Llama3 Swallow Instruct|\n|---|---|---|\n|8B| [Link](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-8B-v0.1) | [Link](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1) |\n|70B| [Link](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-v0.1) | [Link](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-Instruct-v0.1) |\n\n![logo](./logo.png)\n\nThis repository provides large language models developed by [Swallow-LLM](https://swallow-llm.github.io/).\nRead our [blog post](https://zenn.dev/tokyotech_lm/articles/f65989d76baf2c).\n\n## Model Details\n\n* **Model type**: Please refer to [Llama 3 MODEL_CARD](https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md) for details on the model architecture.\n* **Language(s)**: Japanese English\n* **Library**: [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) \n* **Tokenizer**: Please refer to [Llama 3 blog](https://ai.meta.com/blog/meta-llama-3/) for details on the tokenizer.\n* **Contact**: swallow[at]nlp.c.titech.ac.jp \n\n## Model Performance\n\n### Japanese tasks\n\n|Model|Size|JCom.|JEMHopQA|NIILC|JSQuAD|XL-Sum|MGSM|WMT20-en-ja|WMT20-ja-en|JMMLU|JHumanEval|Ja Avg|\n|---|---|---|---|---|---|---|---|---|---|---|---|---|\n|   |   |4-shot|4-shot|4-shot|4-shot|1-shot|4-shot|4-shot|4-shot|5-shot|0-shot|   |\n|   |   |EM acc|Char-F1|Char-F1|Char-F1|ROUGE-2|EM acc|BLEU|BLEU|EM acc|pass@1|   |\n|karakuri-lm-70b-chat-v0.1|70B|0.8847|0.5139|0.5668|0.9096|0.1369|0.2800|0.2526|0.2095|0.4648|0.2354|0.4454|\n|Meta-Llama-3-70B-Instruct|70B|0.9419|0.6114|0.5506|0.9164|0.1912|0.7200|0.2708|0.2350|0.6789|0.6610|0.5777|\n|Llama-3-Swallow-70B-Instruct-v0.1|70B|0.9607|0.6188|0.6026|0.9236|0.1389|0.6560|0.2724|0.2532|0.6572|0.6000|0.5683|\n|Qwen2-72B-Instruct|72B|0.9634|0.6268|0.5418|0.9210|0.1644|0.7840|0.2592|0.2327|0.7713|0.6909|0.5955|\n\n### English tasks\n\n|Model|Size|OpenBookQA|TriviaQA|HellaSWAG|SQuAD2.0|XWINO|MMLU|GSM8K|BBH|HumanEval|EnAvg|\n|---|---|---|---|---|---|---|---|---|---|---|---|\n|||4-shot|4-shot|4-shot|4-shot|4-shot|5-shot|4-shot|3-shot|0-shot||\n|||Acc|EMacc|Acc|EMacc|Acc|Acc|EMacc|CoTEMAcc|pass@1||\n|karakuri-lm-70b-chat-v0.1|70B|0.4100|0.6873|0.6315|0.3677|0.9049|0.5941|0.3882|0.5724|0.2305|0.5319|\n|Meta-Llama-3-70B-Instruct|70B|00.4400|0.7999|0.6552|0.4024|0.9127|0.7992|0.9052|0.8326|0.7555|0.7225|\n|Llama-3-Swallow-70B-Instruct-v0.1|70B|0.4520|0.8174|0.6758|0.4050|0.9230|0.7883|0.8688|0.8152|0.6890|0.7150|\n|Qwen2-72B-Instruct|72B|0.4360|0.7588|0.6857|0.3913|0.9110|0.8391|0.8499|0.2436|0.6939|0.6455|\n\n## MT-Bench JA\n\n|Model|Size|coding|extraction|humanities|math|reasoning|roleplay|stem|writing|JMTAvg|\n|---|---|---|---|---|---|---|---|---|---|---|\n|karakuri-lm-70b-chat-v0.1|70B|0.2804|0.5862|0.6240|0.2934|0.4183|0.5530|0.4859|0.5964|0.4797|\n|Meta-Llama-3-70B-Instruct|70B|0.5969|0.8410|0.7120|0.4481|0.4884|0.7117|0.6510|0.6900|0.6424|\n|Llama-3-Swallow-70B-Instruct-v0.1|70B|0.5269|0.7250|0.5690|0.4669|0.6121|0.6238|0.5533|0.5698|0.5809|\n|Qwen2-72B-Instruct|72B|0.5699|0.7858|0.8222|0.5096|0.7032|0.7963|0.7728|0.8223|0.7228|\n|GPT-3.5(gpt-3.5-turbo-0125)| |0.6851|0.7641|0.7414|0.5522|0.5128|0.7104|0.6266|0.7361|0.6661|\n|GPT-4o(gpt-4o-2024-05-13)| |0.7296|0.8540|0.8646|0.6641|0.6661|0.8274|0.8184|0.8085|0.7791|\n\n## Evaluation Benchmarks\n\n### Japanese evaluation benchmarks\n\nWe 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:\n\n- Multiple-choice question answering (JCommonsenseQA [Kurihara et al., 2022])\n- Open-ended question answering (JEMHopQA [Ishii et al., 2024])\n- Open-ended question answering (NIILC [関根, 2003])\n- Machine reading comprehension (JSQuAD [Kurihara et al., 2022])\n- Automatic summarization (XL-Sum [Hasan et al., 2021])\n- Machine translation (WMT2020 ja-en [Barrault et al., 2020])\n- Machine translation (WMT2020 en-ja [Barrault et al., 2020])\n- Mathematical reasoning (MGSM [Shi et al., 2023])\n- Academic exams (JMMLU [尹ら, 2024])\n- Code generation (JHumanEval [佐藤ら, 2024])\n\n### English evaluation benchmarks\n\nWe used the Language Model Evaluation Harness(v.0.4.2) and Code Generation LM Evaluation Harness(commit #0261c52). The details are as follows:\n\n- Multiple-choice question answering (OpenBookQA [Mihaylov et al., 2018])\n- Open-ended question answering (TriviaQA [Joshi et al., 2017])\n- Machine reading comprehension (SQuAD2 [Rajpurkar et al., 2018])\n- Commonsense reasoning (XWINO [Tikhonov and Ryabinin, 2021])\n- Natural language inference (HellaSwag [Zellers et al., 2019])\n- Mathematical reasoning (GSM8K [Cobbe et al., 2021])\n- Reasoning (BBH (BIG-Bench-Hard) [Suzgun et al., 2023])\n- Academic exams (MMLU [Hendrycks et al., 2021])\n- Code generation (HumanEval [Chen et al., 2021])\n\n### MT-Bench JA\n\nWe used [Japanese MT-Bench](https://wandb.ai/wandb-japan/llm-leaderboard/artifacts/dataset/mtbench_ja_question) to assess the instruction-following capabilities of models.\nWe utilized the following settings:\n\n- Implemantation: FastChat [Zheng+, 2023] (commit #e86e70d0)\n- Question: [Nejumi LLM-Leaderboard NEO, mtbench_ja_question_v3](https://wandb.ai/wandb-japan/llm-leaderboard/artifacts/dataset/mtbench_ja_question/v3)\n- Reference Answer: [Nejumi LLM-Leaderboard NEO, mtbench_ja_referenceanswer_v1](https://wandb.ai/wandb-japan/llm-leaderboard/artifacts/dataset/mtbench_ja_referenceanswer/v1)\n- Prompt for Judge: [Nejumi LLM-Lederboard NEO, mtbench_ja_prompt_v1](https://wandb.ai/wandb-japan/llm-leaderboard/artifacts/dataset/mtbench_ja_prompt/v1)\n- Judge: `gpt-4-1106-preview`\n- Scoring: Absolute scale normalized to a 0-1 range, averaged over five runs.\n\n## Usage\n\n```sh\npip install vllm\n```\n\n```python\nfrom transformers import AutoTokenizer\nfrom vllm import LLM, SamplingParams\n\nmodel_name = \"tokyotech-llm/Llama-3-Swallow-70B-Instruct-v0.1\"\n\ntokenizer = AutoTokenizer.from_pretrained(model_name)\nllm = LLM(\n    model=model_name,\n    tensor_parallel_size=4,\n)\n\nsampling_params = SamplingParams(\n    temperature=0.6, top_p=0.9, max_tokens=512, stop=\"<|eot_id|>\"\n)\n\n\nmessage = [\n    {\"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。\"},\n    {\n        \"role\": \"user\",\n        \"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。\",\n    },\n]\nprompt = tokenizer.apply_chat_template(\n    message, tokenize=False, add_generation_prompt=True\n)\n\noutput = llm.generate(prompt, sampling_params)\n\nprint(output[0].outputs[0].text)\n\n```\n\n## Training Datasets\n\n### Instruction Tuning\n\nThe following datasets were used for the instruction tuning. \n\n- [OpenAssistant Conversations Dataset EN top-1 thread](https://huggingface.co/datasets/OpenAssistant/oasst2)\n- [OpenAssistant Conversations Dataset](https://huggingface.co/datasets/llm-jp/oasst1-21k-ja) 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](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1) model.\n\n \n## Risks and Limitations\n\nThe 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.\n\n## Acknowledgements\n\nWe thank Meta Research for releasing Llama 3 under an open license for others to build on.\n\nOur project is supported by the [Large Generative AI Development Support Program](https://abci.ai/en/link/lfm_support_program.html) of the National Institute of Advanced Industrial Science and Technology. \n\n## License\n\n[META LLAMA 3 COMMUNITY LICENSE](https://llama.meta.com/llama3/license/)\n\n## Authors\n\nHere are the team members:\n- From [Tokyo Institute of Technology Okazaki Laboratory](https://www.nlp.c.titech.ac.jp/index.en.html), the following members:\n  - [Naoaki Okazaki](https://www.chokkan.org/index.ja.html)\n  - [Sakae Mizuki](https://s-mizuki-nlp.github.io/)\n  - [Youmi Ma](https://www.nlp.c.titech.ac.jp/member/youmi.en.html)\n  - [Koki Maeda](https://sites.google.com/view/silviase)\n  - [Kakeru Hattori](https://aya-se.vercel.app/)\n  - [Masanari Ohi](https://sites.google.com/view/masanariohi)\n  - [Taihei Shiotani](https://github.com/inatoihs)\n  - [Koshiro Saito](https://sites.google.com/view/koshiro-saito)\n- From [Tokyo Institute of Technology YOKOTA Laboratory](https://www.rio.gsic.titech.ac.jp/en/index.html), the following members:\n  - [Rio Yokota](https://twitter.com/rioyokota)\n  - [Kazuki Fujii](https://twitter.com/okoge_kaz)\n  - [Taishi Nakamura](https://twitter.com/Setuna7777_2)\n  - [Takumi Okamoto](https://www.linkedin.com/in/takumi-okamoto)\n  - [Ishida Shigeki](https://www.wantedly.com/id/reborn27)\n- From [Artificial Intelligence Research Center, AIST, Japan](https://www.airc.aist.go.jp/en/teams/), the following members:\n  - [Hiroya Takamura](https://sites.google.com/view/hjtakamura)\n\n## How to cite\n\nIf you find our work helpful, please feel free to cite us.\n\n```\n@inproceedings{Fujii:COLM2024,\n   title={Continual Pre-Training for Cross-Lingual LLM Adaptation:\nEnhancing Japanese Language Capabilities},\n   author={Kazuki Fujii and Taishi Nakamura and Mengsay Loem and Hiroki\nIida and Masanari Ohi and Kakeru Hattori and Hirai Shota and Sakae\nMizuki and Rio Yokota and Naoaki Okazaki},\n   booktitle=\"Proceedings of the First Conference on Language Modeling\",\n   series={COLM},\n   pages=\"(to appear)\",\n   year=\"2024\",\n   month=oct,\n   address={University of Pennsylvania, USA},\n}\n\n@inproceedings{Okazaki:COLM2024,\n   title={Building a Large Japanese Web Corpus for Large Language Models},\n   author={Naoaki Okazaki and Kakeru Hattori and Hirai Shota and Hiroki\nIida and Masanari Ohi and Kazuki Fujii and Taishi Nakamura and Mengsay\nLoem and Rio Yokota and Sakae Mizuki},\n   booktitle=\"Proceedings of the First Conference on Language Modeling\",\n   series={COLM},\n   pages=\"(to appear)\",\n   year=\"2024\",\n   month=oct,\n   address={University of Pennsylvania, USA},\n}\n```\n\n### Citations\n\n```tex\n@article{llama3modelcard,\n    title={Llama 3 Model Card},\n    author={AI@Meta},\n    year={2024},\n    url = {https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md}\n}\n```\n\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 0,
  "downloads": 659,
  "gated": false,
  "private": false,
  "last_modified": "2024-09-08T02:02:10.000Z",
  "created_at": "2024-09-07T05:16:12.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "66dbe19cfc68bfbb93b5041a",
  "id": "RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf",
  "modelId": "RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-Instruct-v0.1-gguf",
  "sha": "ab2a57c2340fcaf6cdc15c904fb20372283eb11f",
  "createdAt": "2024-09-07T05:16:12.000Z",
  "lastModified": "2024-09-08T02:02:10.000Z",
  "author": "RichardErkhov",
  "downloads": 659,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 36
}