模型简介
模型特点
模型能力
使用案例
🚀 仙都(Cendol):面向印尼语的开源指令调优生成式大语言模型
仙都(Cendol)是一个开源的、针对印尼语进行微调的生成式大语言模型集合,涵盖了仅解码器和编码器 - 解码器两种Transformer模型架构,参数规模从3亿到130亿不等。
本仓库为 70亿参数的仙都(Cendol)LLaMA - 2聊天模型。其他模型的链接如下文所示。
✨ 主要特性
- 多架构与多参数规模:基于mT5和LLaMA - 2两种基础模型,提供多种参数规模的模型,满足不同场景需求。
- 指令调优版本丰富:包括针对特定任务的Cendol - Instruct和基于通用知识与以人为中心提示进行持续调优的Cendol - Chat,均适用于单轮对话。
- 性能卓越:在大多数测试基准上,大幅超越开源的多语言和特定地区大语言模型,小版本(参数少于10亿)也能与70亿参数的其他模型相媲美。
- 高效策略:提出词汇替换策略,相比LoRA调优,在训练和推理时间上分别提高了11.50%和18.71%,且评估性能与使用原始词汇训练的模型相当。
📚 详细文档
模型详情
- 注意:仙都(Cendol)的使用遵循 [Apache 2.0许可证](https://choosealicense.com/licenses/apache - 2.0/)。
- 概述:由IndoNLP开发并公开发布,是一系列预训练和微调的生成式文本模型,参数规模从5.6亿到130亿不等。
- 模型开发者:IndoNLP
- 变体:基于mT5的模型有3亿(mT5 - small)、5.8亿(mT5 - base)、12亿(mT5 - large)、37亿(mT5 - XL)和130亿(mT5 - XXL);基于LLaMA - 2的模型有70亿(LLaMA2 - 7B)和130亿(LLaMA2 - 13B)。均有Cendol - Instruct和Cendol - Chat两种变体。130亿参数的模型采用LoRA调优,其他则进行全量微调。
- 输入输出:模型的输入和输出均为文本。
- 模型架构
模型 | 训练数据 | 参数 | 调优策略 | 学习率 |
---|---|---|---|---|
[仙都(Cendol)mT5 - small Instruct](https://huggingface.co/indonlp/cendol - mT5 - small - inst) | 仙都(Cendol)集合v1 | 3亿 | 全量微调 | 3.0 x 10⁻⁴ |
[仙都(Cendol)mT5 - base Instruct](https://huggingface.co/indonlp/cendol - mT5 - base - inst) | 仙都(Cendol)集合v1 | 5.8亿 | 全量微调 | 3.0 x 10⁻⁴ |
[仙都(Cendol)mT5 - large Instruct](https://huggingface.co/indonlp/cendol - mT5 - large - inst) | 仙都(Cendol)集合v1 | 12亿 | 全量微调 | 3.0 x 10⁻⁴ |
[仙都(Cendol)mT5 - XL Instruct](https://huggingface.co/indonlp/cendol - mT5 - XL - inst) | 仙都(Cendol)集合v1 | 37亿 | 全量微调 | 3.0 x 10⁻⁴ |
[仙都(Cendol)mT5 - XXL Instruct](https://huggingface.co/indonlp/cendol - mT5 - XXL - merged - inst) | 仙都(Cendol)集合v1 | 130亿 | LoRA | 2.0 x 10⁻⁴ |
[仙都(Cendol)LLaMA - 2 (7B) Instruct](https://huggingface.co/indonlp/cendol - llama2 - 7B - inst) | 仙都(Cendol)集合v1 | 70亿 | 全量微调 | 2.0 x 10⁻⁵ |
[仙都(Cendol)LLaMA - 2 (7B) 印尼语词汇指令调优模型](https://huggingface.co/indonlp/cendol - llama2 - ind - vocab - inst) | 仙都(Cendol)集合v1 | 70亿 | 全量微调 | 2.0 x 10⁻⁵ |
[仙都(Cendol)LLaMA - 2 (13B) Instruct](https://huggingface.co/indonlp/cendol - llama2 - 13B - merged - inst) | 仙都(Cendol)集合v1 | 130亿 | LoRA | 2.0 x 10⁻⁵ |
[仙都(Cendol)mT5 - small Chat](https://huggingface.co/indonlp/cendol - mT5 - small - chat) | 仙都(Cendol)集合v2 | 3亿 | 全量微调 | 3.0 x 10⁻⁵ |
[仙都(Cendol)mT5 - base Chat](https://huggingface.co/indonlp/cendol - mT5 - base - chat) | 仙都(Cendol)集合v2 | 5.8亿 | 全量微调 | 3.0 x 10⁻⁵ |
[仙都(Cendol)mT5 - large Chat](https://huggingface.co/indonlp/cendol - mT5 - large - chat) | 仙都(Cendol)集合v2 | 12亿 | 全量微调 | 3.0 x 10⁻⁵ |
[仙都(Cendol)mT5 - XL Chat](https://huggingface.co/indonlp/cendol - mT5 - XL - chat) | 仙都(Cendol)集合v2 | 37亿 | 全量微调 | 3.0 x 10⁻⁵ |
[仙都(Cendol)mT5 - XXL Chat](https://huggingface.co/indonlp/cendol - mT5 - XXL - merged - chat) | 仙都(Cendol)集合v2 | 130亿 | LoRA | 2.0 x 10⁻⁴ |
[仙都(Cendol)LLaMA - 2 (7B) Chat](https://huggingface.co/indonlp/cendol - llama2 - 7B - chat) | 仙都(Cendol)集合v2 | 70亿 | 全量微调 | 1.0 x 10⁻⁵ |
[仙都(Cendol)LLaMA - 2 (13B) Chat](https://huggingface.co/indonlp/cendol - llama2 - 13B - merged - chat) | 仙都(Cendol)集合v2 | 130亿 | LoRA | 2.0 x 10⁻⁴ |
- 模型训练时间:仙都(Cendol)于2023年10月至2024年1月期间进行训练。
- 许可证:使用仙都(Cendol)遵循 [Apache 2.0许可证](https://choosealicense.com/licenses/apache - 2.0/)
- 研究论文:"仙都(Cendol):面向印尼语的开源指令调优生成式大语言模型"
预期用途
- 预期用例:仙都(Cendol)主要用于研究,特别是针对印尼语的研究。Cendol - Instruct模型可用于特定任务指令,Cendol - Chat模型可用于通用知识指令。
- 非预期用途:禁止以任何违反适用法律法规(包括贸易合规法律)的方式使用;禁止用于英语和印尼语以外的语言;禁止以《仙都(Cendol)可接受使用政策和许可协议》禁止的任何其他方式使用。
评估结果
在本节中,我们报告了仙都(Cendol)模型在大规模自然语言理解(NLU)和自然语言生成(NLG)基准测试中的结果。所有评估均使用我们的内部评估库。
NLU性能

NLG性能

人工评估

伦理考量与局限性
仙都(Cendol)是一项新技术,使用时存在风险。到目前为止的测试均使用印尼语进行,无法涵盖所有场景。因此,与所有大语言模型一样,仙都(Cendol)的潜在输出无法提前预测,在某些情况下,模型可能会对用户提示产生不准确、有偏见或其他令人反感的回复。所以,在部署仙都(Cendol)的任何应用之前,开发者应针对其特定应用对模型进行安全测试和调优。
引用
如果您使用了包括仙都(Cendol)模型、代码或数据在内的任何资源,请引用以下文章:
@misc{cahyawijaya - etal - 2024 - cendol,
title={Cendol: Open Instruction - tuned Generative Large Language Models for Indonesian Languages},
author={Samuel Cahyawijaya and Holy Lovenia and Fajri Koto and Rifki Afina Putri and Emmanuel Dave and Jhonson Lee and Nuur Shadieq and Wawan Cenggoro and Salsabil Maulana Akbar and Muhammad Ihza Mahendra and Dea Annisayanti Putri and Bryan Wilie and Genta Indra Winata and Alham Fikri Aji and Ayu Purwarianti and Pascale Fung},
year={2024},
eprint={2404.06138},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
@inproceedings{cahyawijaya - etal - 2023 - nusacrowd,
title = "{N}usa{C}rowd: Open Source Initiative for {I}ndonesian {NLP} Resources",
author = "Cahyawijaya, Samuel and
Lovenia, Holy and
Aji, Alham Fikri and
Winata, Genta and
Wilie, Bryan and
Koto, Fajri and
Mahendra, Rahmad and
Wibisono, Christian and
Romadhony, Ade and
Vincentio, Karissa and
Santoso, Jennifer and
Moeljadi, David and
Wirawan, Cahya and
Hudi, Frederikus and
Wicaksono, Muhammad Satrio and
Parmonangan, Ivan and
Alfina, Ika and
Putra, Ilham Firdausi and
Rahmadani, Samsul and
Oenang, Yulianti and
Septiandri, Ali and
Jaya, James and
Dhole, Kaustubh and
Suryani, Arie and
Putri, Rifki Afina and
Su, Dan and
Stevens, Keith and
Nityasya, Made Nindyatama and
Adilazuarda, Muhammad and
Hadiwijaya, Ryan and
Diandaru, Ryandito and
Yu, Tiezheng and
Ghifari, Vito and
Dai, Wenliang and
Xu, Yan and
Damapuspita, Dyah and
Wibowo, Haryo and
Tho, Cuk and
Karo Karo, Ichwanul and
Fatyanosa, Tirana and
Ji, Ziwei and
Neubig, Graham and
Baldwin, Timothy and
Ruder, Sebastian and
Fung, Pascale and
Sujaini, Herry and
Sakti, Sakriani and
Purwarianti, Ayu",
editor = "Rogers, Anna and
Boyd - Graber, Jordan and
Okazaki, Naoaki",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.findings - acl.868",
doi = "10.18653/v1/2023.findings - acl.868",
pages = "13745--13818"
}
此外,如果您受到我们针对印尼语及其当地语言的特定地区大语言模型研究工作的启发,请考虑引用以下文章:
@inproceedings{cahyawijaya - etal - 2023 - nusawrites,
title = "{N}usa{W}rites: Constructing High - Quality Corpora for Underrepresented and Extremely Low - Resource Languages",
author = "Cahyawijaya, Samuel and
Lovenia, Holy and
Koto, Fajri and
Adhista, Dea and
Dave, Emmanuel and
Oktavianti, Sarah and
Akbar, Salsabil and
Lee, Jhonson and
Shadieq, Nuur and
Cenggoro, Tjeng Wawan and
Linuwih, Hanung and
Wilie, Bryan and
Muridan, Galih and
Winata, Genta and
Moeljadi, David and
Aji, Alham Fikri and
Purwarianti, Ayu and
Fung, Pascale",
editor = "Park, Jong C. and
Arase, Yuki and
Hu, Baotian and
Lu, Wei and
Wijaya, Derry and
Purwarianti, Ayu and
Krisnadhi, Adila Alfa",
booktitle = "Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia - Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = nov,
year = "2023",
address = "Nusa Dua, Bali",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.ijcnlp - main.60",
doi = "10.18653/v1/2023.ijcnlp - main.60",
pages = "921--945"
}
@inproceedings{winata - etal - 2023 - nusax,
title = "{N}usa{X}: Multilingual Parallel Sentiment Dataset for 10 {I}ndonesian Local Languages",
author = "Winata, Genta Indra and
Aji, Alham Fikri and
Cahyawijaya, Samuel and
Mahendra, Rahmad and
Koto, Fajri and
Romadhony, Ade and
Kurniawan, Kemal and
Moeljadi, David and
Prasojo, Radityo Eko and
Fung, Pascale and
Baldwin, Timothy and
Lau, Jey Han and
Sennrich, Rico and
Ruder, Sebastian",
editor = "Vlachos, Andreas and
Augenstein, Isabelle",
booktitle = "Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics",
month = may,
year = "2023",
address = "Dubrovnik, Croatia",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.eacl - main.57",
doi = "10.18653/v1/2023.eacl - main.57",
pages = "815--834"
}
@inproceedings{aji - etal - 2022 - one,
title = "One Country, 700 + Languages: {NLP} Challenges for Underrepresented Languages and Dialects in {I}ndonesia",
author = "Aji, Alham Fikri and
Winata, Genta Indra and
Koto, Fajri and
Cahyawijaya, Samuel and
Romadhony, Ade and
Mahendra, Rahmad and
Kurniawan, Kemal and
Moeljadi, David and
Prasojo, Radityo Eko and
Baldwin, Timothy and
Lau, Jey Han and
Ruder, Sebastian",
editor = "Muresan, Smaranda and
Nakov, Preslav and
Villavicencio, Aline",
booktitle = "Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = may,
year = "2022",
address = "Dublin, Ireland",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.acl - long.500",
doi = "10.18653/v1/2022.acl - long.500",
pages = "7226--7249"
}
@inproceedings{cahyawijaya - etal - 2021 - indonlg,
title = "{I}ndo{NLG}: Benchmark and Resources for Evaluating {I}ndonesian Natural Language Generation",
author = "Cahyawijaya, Samuel and
Winata, Genta Indra and
Wilie, Bryan and
Vincentio, Karissa and
Li, Xiaohong and
Kuncoro, Adhiguna and
Ruder, Sebastian and
Lim, Zhi Yuan and
Bahar, Syafri and
Khodra, Masayu and
Purwarianti, Ayu and
Fung, Pascale",
editor = "Moens, Marie - Francine and
Huang, Xuanjing and
Specia, Lucia and
Yih, Scott Wen - tau",
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2021",
address = "Online and Punta Cana, Dominican Republic",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.emnlp - main.699",
doi = "10.18653/v1/2021.emnlp - main.699",
pages = "8875--8898"
}
@inproceedings{wilie - etal - 2020 - indonlu,
title = "{I}ndo{NLU}: Benchmark and Resources for Evaluating {I}ndonesian Natural Language Understanding",
author = "Wilie, Bryan and
Vincentio, Karissa and
Winata, Genta Indra and
Cahyawijaya, Samuel and
Li, Xiaohong and
Lim, Zhi Yuan and
Soleman, Sidik and
Mahendra, Rahmad and
Fung, Pascale and
Bahar, Syafri and
Purwarianti, Ayu",
editor = "Wong, Kam - Fai and
Knight, Kevin and
Wu, Hua",
booktitle = "Proceedings of the 1st Conference of the Asia - Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing",
month = dec,
year = "2020",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2020.aacl - main.85",
pages = "843--857"
}
📄 许可证
仙都(Cendol)的使用遵循 [Apache 2.0许可证](https://choosealicense.com/licenses/apache - 2.0/)。



