文章主题：moss, moon, plugin
IT之家 4 月 21 日消息，复旦大学自然语言处理实验室开发的新版 MOSS 模型今日正式上线，成为国内首个插件增强的开源对话语言模型。
目前，MOSS 模型已上线开源，相关代码、数据、模型参数已在 Github 和 Hugging Face 等平台开放，供科研人员下载。
据介绍，MOSS 是一个支持中英双语和多种插件的开源对话语言模型，moss-moon 系列模型具有 160 亿参数，在 FP16 精度下可在单张 A100 / A800 或两张 3090 显卡运行，在 INT4/8 精度下可在单张 3090 显卡运行。MOSS 基座语言模型在约七千亿中英文以及代码单词上预训练得到，后续经过对话指令微调、插件增强学习和人类偏好训练具备多轮对话能力及使用多种插件的能力。
MOSS 来自复旦大学自然语言处理实验室的邱锡鹏教授团队，名字与《流浪地球》电影中的 AI 同名，已发布至公开平台（https://moss.fastnlp.top/），邀请公众参与内测。
IT之家查看 MOSS 的 GitHub 页面发现，该项目所含代码采用 Apache 2.0 协议，数据采用 CC BY-NC 4.0 协议，模型权重采用 GNU AGPL 3.0 协议。如需将该项目所含模型用于商业用途或公开部署，需要签署文件并发送至 robot@fudan.edu.cn 取得授权，商用情况仅用于记录，不会收取任何费用。
MOSS 用例：
🌟 Moss-Moon-003 Base: MOSS-003 Foundation Model 🌟🚀 This foundation model is trained on high-quality Chinese and English datasets using self-supervised pre-training, with approximately 700 billion words of data and 6.67×1022 floating-point operations.💡 The pre-trained dataset includes a diverse range of text genres to ensure the model’s versatility in various applications.🌐 This foundation model is available for research purposes only, and its use must be strictly adhered to all relevant laws and regulations.📝 For more information on this model or any other Moss-Moon product, please contact our customer support team at [insert contact details here].# Moss-Moon-003 Base 🌟 Foundation Model🚀 This foundation model is trained on high-quality Chinese and English datasets using self-supervised pre-training, with approximately 700 billion words of data and 6.67×1022 floating-point operations.💡 The pre-trained dataset includes a diverse range of text genres to ensure the model’s versatility in various applications.🌐 This foundation model is available for research purposes only, and its use must be strictly adhered to all relevant laws and regulations.📝 For more information on this model or any other Moss-Moon product, please contact our customer support team at [insert contact details here].
moss-moon-003-sft: 基座模型在约 110 万多轮对话数据上微调得到，具有指令遵循能力、多轮对话能力、规避有害请求能力。
🌟 Moss Moon 003 SFT Plugin: A Foundation Model Derived from Over 110,000 Round Conversations and Approximately 30,000 Plugin Enhancements 🌟The foundation model was fine-tuned using over 110,000 rounds of conversation data and approximately 30,000 plugin enhancements. It also has the ability to utilize search engines, generate text from images, perform calculations, and solve equations.💡 Moss Moon 003 SFT Plugin: A Foundation Model Derived from Over 110,000 Round Conversations and Approximately 30,000 Plugin Enhancements 📈The foundation model was refined using over 110,000 conversations and approximately 30,000 plugin enhancements. It also has the capability to utilize search engines, generate text from images, perform calculations, and solve equations.💡 Moss Moon 003 SFT Plugin: A Foundation Model Derived from Over 110,000 Round Conversations and Approximately 30,000 Plugin Enhancements 📈The foundation model was optimized using over 110,000 conversations and approximately 30,000 plugin enhancements. It also has the ability to utilize search engines, generate text from images, perform calculations, and solve equations.💡 Moss Moon 003 SFT Plugin: A Foundation Model Derived from Over 110,000 Round Conversations and Approximately 30,000 Plugin Enhancements 📈The foundation model was refined using over 110,000 conversations and approximately 30,000 plugin enhancements. It also has the ability to utilize search engines, generate text from images, perform calculations, and solve equations.💡 Moss Moon 003 SFT Plugin: A Foundation Model Derived from Over 110,000 Round Conversations and Approximately 30,000 Plugin Enhancements 📈The foundation model was optimized using over 110,000 conversations and approximately 30,000 plugin enhancements. It also has the ability to utilize search engines, generate text from images, perform calculations, and solve equations.💡 Moss Moon 003 SFT Plugin: A Foundation Model Derived from Over 110,000 Round Conversations and Approximately 30,000 Plugin Enhancements 📈The foundation model was refined using over 110,000 conversations and approximately 30,000 plugin enhancements. It also has the ability to utilize search engines, generate text from images, perform calculations, and solve equations.💡 Moss Moon 003 SFT Plugin: A Foundation Model Derived from Over 110,000 Round Conversations and Approximately 30,000 Plugin Enhancements 📈
🎉 Moss Moon 003 PM 🎉🚀 We are excited to announce that our preference model trained on the data collected by Moss Moon 003 SFT will be released soon! 🚀💡 Our team has put in a lot of effort to create an accurate and reliable preference model. This model is based on the feedback we received from users like you, which helped us understand your needs better.🚀 We are committed to making our research accessible to everyone. That’s why we plan to release this model soon so that you can benefit from it too.💡 If you have any questions or suggestions about the model, feel free to reach out to us. We would love to hear from you!🎉 Stay tuned for more updates on Moss Moon 003 PM! 🎉#MossMoon #PreferenceModel #OpenSource #ResearchAccessibility
🚀 Moss Moon 003: The Final Model 🚀🚀 Moss Moon 003 is the result of a comprehensive training process on the Moss Moon 003-SFT foundation, resulting in a model that boasts superior factual accuracy and security. Its stable response quality ensures that users can rely on it for reliable information.🚀 We are excited to announce that this advanced model will soon be made available for public use. Stay tuned for updates on its release date!#MossMoon003 #FinalModel #FactualAccuracy #Security #StableResponseQuality #PublicUse #ReleaseDate #StayTuned
🚀 Moss Moon 003 Plugin: A Stronger Intent Understanding and Plugin Usage Capable Model Trained on Moss Moon 003 SFT Plugin’s Foundation 🚀🚀 Moss Moon 003 Plugin is a powerful tool that has been trained using the Moss Moon 003 SFT Plugin as its foundation. This model has undergone extensive training to enhance its ability to understand user intent and make more informed decisions when it comes to plugin usage.🚀 The Moss Moon 003 PM, which stands for Preference Model, plays a crucial role in this process. It helps the model learn from user behavior and preferences, allowing it to adapt to different situations and provide better results.🚀 With its improved capabilities, the Moss Moon 003 Plugin is set to be released soon. This exciting development will enable users to take advantage of the latest advancements in plugin usage and intent understanding.🚀 So stay tuned for more updates on this powerful tool! 🚀# moss-moon-003-plugin #intent-understanding #plugin-usage-capable-model #training #moss-moon-003-sft-plugin #preference-model #release #exciting-development #user-behavior #preferences #latest-advancements #take-advantage #upcoming-release #stay-tuned
数据# Moss-002 SFT DataMoss-002 uses multi-round dialogue data to cover the three aspects of usefulness, loyalty, and harmlessness. It contains approximately 570,000 English conversations generated by text-davinci-003 and 590,000 Chinese conversations.
moss-003-sft-data: moss-moon-003-sft 所使用的多轮对话数据，基于 MOSS-002 内测阶段采集的约 10 万用户输入数据和 gpt-3.5-turbo 构造而成，相比 moss-002-sft-data，moss-003-sft-data 更加符合真实用户意图分布，包含更细粒度的有用性类别标记、更广泛的无害性数据和更长对话轮数，约含 110 万条对话数据。目前仅开源少量示例数据，完整数据将在近期开源。
moss-003-sft-plugin-data: moss-moon-003-sft-plugin 所使用的插件增强的多轮对话数据，包含支持搜索引擎、文生图、计算器、解方程等四个插件在内的约 30 万条多轮对话数据。目前仅开源少量示例数据，完整数据将在近期开源。
moss-003-pm-data: moss-moon-003-pm 所使用的偏好数据，包含在约 18 万额外对话上下文数据及使用 moss-moon-003-sft 所产生的回复数据上构造得到的偏好对比数据，将在近期开源。
MOSS 的 GitHub 页面：点此查看
AI时代，掌握AI大模型第一手资讯！AI时代不落人后！
免费ChatGPT问答，办公、写作、生活好得力助手！
扫码右边公众号，驾驭AI生产力！
