GPT Agents Products
icon
password
Collection of Open sourced GPT agents products。
- Adala - An autonomous data (labeling) agent framework that is reliable, controllable, and specialized in data processing.
- Agent4Rec - A recommender system simulator using 1,000 LLM-empowered generative agents, initialized from the MovieLens-1M dataset.
- AgentForge - A low-code framework for creating, testing, and iterating AI-powered autonomous agents and cognitive architectures, compatible with various LLM models.
- AgentGPT - A browser-based implementation of AutoGPT, accessible via a no-code platform.
- Agents - An open-source library/framework for building autonomous language agents with features like long-short term memory, tool usage, web navigation, multi-agent communication, human-agent interaction, and symbolic control.
- AgentVerse - A platform for deploying multiple LLM-based agents in various applications, focusing on task-solving and simulation frameworks.
- AI Legion - A platform for LLM-powered autonomous agents to work together and accomplish tasks.
- Aider - A command-line tool for pairing programming with GPT-3.5/GPT-4, editing code stored in local git repositories.
- AutoGen - A framework for developing LLM applications using multiple conversational agents that can collaborate to solve tasks and interact with humans.
- AutoGPT - An open-source attempt to make GPT-4 fully autonomous, capable of chaining together LLM "thoughts" to autonomously achieve goals.
- Automata - A system that autonomously generates its own code based on the context of your project, aiming to evolve into a fully autonomous, self-programming AI system.
- AutoPR - AI-generated pull requests to fix issues, powered by ChatGPT.
- Autonomous HR Chatbot - A prototype enterprise application, an autonomous HR assistant powered by GPT-3.5.
- BabyAGI - A framework for managing tasks using AI, creating tasks based on previous results and predefined objectives.
- BabyBeeAGI - A task management and functionality expansion on top of BabyAGI, designed to handle multiple functions within one task management prompt.
- BabyCatAGI - A mod of BabyBeeAGI, focusing on task creation and execution with additional tools.
- BabyDeerAGI - A mod of BabyAGI, featuring parallel tasks and user input tools.
- BabyElfAGI - The newest mod of BabyDeerAGI and OG BabyAGI, featuring skills classes, dynamic task lists, and a beta reflection agent.
- BabyCommandAGI - An AI agent that combines CLI and LLM, executing shell commands and automating tasks like app development and environment setup.
- BabyFoxAGI - A mod of BabyElfAGI with self-improving task lists and a novel chat UI with parallel tasks.
- BambooAI - A semi-autonomous AI agent for data exploration and analysis, designed for non-programmers.
- BeeBot - An early-stage research project focused on developing an autonomous AI assistant for practical tasks.
- Bloop - A GPT-4 powered semantic code search engine with AI assistance for precise code navigation.
- BondAI - A versatile autonomous AI agent with CLI, API, and Docker image, integrating various tools and applications.
- Cal.ai - An open-source scheduling assistant built on Cal.com, capable of booking meetings and summarizing schedules using natural language.
- CAMEL - A platform for studying autonomous and communicative agents, featuring AI user and assistant agents.
- ChatDev - A virtual software company simulation with agents assuming different roles in software development.
- ChemCrow - A chemistry agent for tasks in synthesis, drug discovery, and materials design, integrating expert-designed tools with LLMs.
- Clippy - An assistant for coding tasks, capable of planning, writing, debugging, and testing projects autonomously.
- Cody by ajhous44 - An AI assistant for interactively querying codebases using natural language, enhancing code navigation.
- Cody by Sourcegraph - A code assistant that writes code and answers questions by reading entire codebases and the code graph.
- Continue - An open-source autopilot for software development, integrating ChatGPT with VS Code for various coding tasks.
- Cursor - An AI-first code editor designed for pair-programming with AI, enhancing software development efficiency.
- Databerry - A no-code platform for creating AI chatbots trained on user-provided data, offering various integrations and features.
- Deepnote AI Copilot: An AI code copilot offering efficient and contextual code suggestions for data exploration in notebooks.
- DemoGPT: Enables quick demo generation for LLM-based applications using prompts.
- DevGPT: A virtual team of developers creating microservices based on AI.
- DevOpsGPT: An AI-driven software development automation solution.
- dotagent: An agent management system for robust AI applications and experimental autonomous agents.
- English Compiler: AI-based compiler for converting English-based markdown specs into functional code.
- Friday: A developer assistant for building Node.js projects with unlimited prompts.
- GeniA: An AI platform for engineering teams, executing tasks in dev & cloud environments.
- Godmode: Inspired by Auto-GPT and BabyAGI, conducts various tasks through a user interface.
- GPT Engineer: An AI agent that generates entire codebases based on prompts.
- GPT Migrate: Facilitates migration of codebases between frameworks or languages.
- GPT Pilot: Writes scalable apps while the developer oversees the implementation.
- GPT Researcher: Generates detailed, factual research reports on any topic.
- GPT Runner: Manages and runs AI presets, allowing conversations with selected files.
- IX: A platform for building, debugging, and deploying collaborative agents and cognitive workflows.
- JARVIS: Connects LLMs with the machine learning community for task planning and execution.
- Langroid - A Python framework for building LLM-powered applications. It is intuitive, lightweight, and extensible, supporting a multi-agent, multi-LLM system. It includes specialized agents like
DocChatAgent
,SQLChatAgent
, andTableChatAgent
.
- Lemon Agent - A Plan-Validate-Solve (PVS) agent for workflow automation. It includes a Planner Agent and a Solver Agent for accurate and reliable operations on various tools like GitHub, HubSpot, or Airtable.
- LLM Agents - A minimalistic library for building agents that leverage large language models to automate tasks through a loop of commands and tool integrations.
- Local GPT - An initiative allowing conversation with documents locally on a device, ensuring 100% privacy as no data leaves the device.
- Loop GPT - A re-implementation of Auto-GPT as a modular and extensible Python package, focusing on minimal prompt overhead and human-in-the-loop feedback.
- LocalGPT - Inspired by privateGPT, this project allows using one's documents as an information source, running on local devices with Vicuna-7B model and InstructorEmbeddings.
- MemGPT - A system that manages different memory tiers in LLMs to provide extended context within the LLM's limited context window, enabling chat with data and creating perpetual chatbots.
- Mentat - An AI tool for coding assistance, coordinating edits across multiple locations and files, unlike Copilot and ChatGPT, with contextual project understanding.
- MetaGPT - A multi-agent framework that takes a one-line requirement and outputs various software development aspects. It assigns different roles to GPTs, forming a collaborative software entity.
- Mini AGI - A minimal general-purpose autonomous agent based on GPT-3.5 / GPT-4, capable of performing various tasks like analyzing stock prices and ordering pizza.
- Multiagent Debate - An implementation of a paper on improving factuality and reasoning in LLMs through multiagent debate.
- Multi GPT - An experimental multi-agent system where multiple "expertGPTs" collaborate to perform tasks, each with their own memory and communication abilities.
- OpenAgents - An open platform for language agents, offering a user-centric, fully open-sourced, and extensible environment for researchers, developers, and general users.
- OpenAGI - An AGI R&D platform enabling agents for both benchmark and open-ended tasks, supporting various language models and multi-modality tool learning.
- Open Interpreter - An open-source interpreter allowing LLMs to run code on a computer to complete tasks, working locally for secure document interactions.
- Pezzo - A development toolkit streamlining prompt design, version management, publishing, collaboration, troubleshooting, and observability for AI prompts.
- Private GPT - A tool for private interaction with documents, functioning offline and built with LangChain, GPT4All, LLaMA, Chroma, and SentenceTransformers.
- PromethAI - A personalized AI assistant that assists with nutrition and other goals, decomposing problems and offering solutions with automated flows.
- React Agent - An open-source React.js autonomous LLM agent designed to generate and compose React components from user stories.
- Smol developer - An AI junior developer that specializes in coherent whole program synthesis, debugging, and decompiling, leveraging GPT-4 for code generation.
- Superagent - A tool for creating AI agents without coding, simplifying the configuration and deployment of LLM Agents to production.
- SuperAGI - An open-source framework for developing and deploying autonomous AI agents with features like agent trajectory fine-tuning and performance telemetry.
- Suspicion Agent - An agent that applies GPT-4's learned knowledge to play imperfect information games with theory of mind awareness.
- Sweep - A GitHub assistant that helps fix small bugs and implement small features, automating code development and debugging tasks.
- Taxy AI - A full browser automation tool that uses GPT-4 to control the browser and perform repetitive actions on behalf of users.
- Teenage AGI - An AI agent inspired by BabyAGI with infinite memory recall, thoughtful action planning, and memory persistence.
- Westworld simulation - A multi-agent simulation library for simulating and optimizing systems with multiple interacting agents, inspired by Unity software.
- Voyager - A LLM-powered autonomous agent in Minecraft, capable of continuous exploration and skill acquisition without human intervention.
- WorkGPT - A GPT agent framework for invoking APIs, designed to automate tasks through a series of conversations with AI until the directive is complete.
- XAgent - An experimental autonomous agent driven by Large Language Models (LLM) that can autonomously solve various tasks.
- yAgents - An Agent-Builder Agent for designing, coding, and debugging AI-powered tools and agents, accessible to users of any technical background.
- Yourgoal - A Swift implementation of BabyAGI, an AI-powered task management system using OpenAI and Pinecone APIs.
- AI-Assisted Development and Code Generation:
- Includes: DevGPT, DevOpsGPT, Friday, GPT Engineer, GPT Migrate, GPT Pilot, English Compiler, Smol developer, Sweep, Taxy AI, WorkGPT, yAgents, Yourgoal.
- AI Research, Analysis, and Educational Tools:
- Includes: GPT Researcher, Deepnote AI Copilot, DemoGPT, Bloop (Code Search), ChemCrow (Chemistry).
- Agent-Based Systems, Frameworks, and Management Tools:
- Includes: Agent4Rec, Agents, AgentVerse, AI Legion, BabyAGI, BabyBeeAGI, BabyCatAGI, BabyDeerAGI, BabyElfAGI, BabyFoxAGI, dotagent, IX, JARVIS, Superagent, SuperAGI, Teenage AGI, XAgent.
- AI Applications, Platforms, and Task-Specific Solutions:
- Includes: AgentGPT, AutoGen, Autonomous HR Chatbot, AutoPR, GeniA, Godmode, GPT Runner, BambooAI, BeeBot, BondAI, Cal.ai, CAMEL, ChatDev, Databerry, Clippy, Cody by ajhous44, Cody by Sourcegraph, Continue, Cursor.
- Advanced AI Applications, Simulations, and Experimental Projects:
- Includes: Suspicion Agent, Westworld simulation, Voyager, MetaGPT, Mini AGI.
- AI Frameworks, Libraries, and Privacy-Focused Computing:
- Includes: Langroid, LLM Agents, LocalGPT, Loop GPT, MemGPT, Mini AGI, Multi GPT, OpenAGI, Private GPT.
- Collaborative, Multi-Agent, and Intelligent Systems:
- Includes: Multiagent Debate, MetaGPT, Lemon Agent, OpenAgents, Pezzo, PromethAI, React Agent.
- AI-Enhanced Development and Collaboration Tools:
- Includes: Mentat, Open Interpreter, Collaborative and Multi-Agent Systems (Multiagent Debate, Multi GPT, MetaGPT), AI-Powered Tools and Applications (Pezzo, PromethAI, React Agent)【15†source】【7†source】.
Chinese Version
开源项目分类1 - On Generative Agents in Recommendation核心功能和特点实验环境与设置结果查看与成本2 - AgentVerse特点和功能实验环境设置场景1 - 模拟环境场景2 - 任务解决本地使用3 - Communicative Agents for Software Development近期推出了多个新功能安装构建运行
开源项目
- Adala - 一个可靠、可控且专注于数据处理的自主数据(标注)代理框架。
- Agent4Rec - 使用1,000个LLM赋能的生成代理的推荐系统模拟器,基于MovieLens-1M数据集初始化。
- AgentForge - 一个低代码框架,用于创建、测试和迭代AI驱动的自主代理和认知架构,兼容各种LLM模型。
- AgentGPT - 基于浏览器的AutoGPT实现,可通过无代码平台访问。
- Agents - 一个用于构建自主语言代理的开源库/框架,具有长短期记忆、工具使用、网络导航、多代理通信、人机交互和符号控制等功能。
- AgentVerse - 一个部署多个基于LLM代理的平台,专注于任务解决和模拟框架。
- AI Legion - 一个LLM驱动的自主代理合作完成任务的平台。
- Aider - 一个命令行工具,用于与GPT-3.5/GPT-4进行配对编程,编辑存储在本地git仓库中的代码。
- AutoGen - 使用多个会话代理开发LLM应用的框架,这些代理可以协作解决任务并与人类互动。
- AutoGPT - 一项使GPT-4完全自主的开源尝试,能够将LLM“思维”链接在一起以自主实现目标。
- Automata - 一个根据项目上下文自主生成代码的系统,旨在进化成为一个完全自主的自编程AI系统。
- AutoPR - 由ChatGPT驱动的AI生成的修复问题的Pull Request。
- Autonomous HR Chatbot - 一个原型企业应用程序,一个由GPT-3.5驱动的自主HR助理。
- BabyAGI - 使用AI管理任务的框架,根据先前结果和预定义目标创建任务。
- BabyBeeAGI - BabyAGI的任务管理和功能扩展,旨在处理一个任务管理提示中的多个功能。
- BabyCatAGI - BabyBeeAGI的修改版,专注于任务创建和执行,并配备额外工具。
- BabyDeerAGI - BabyAGI的修改版,具有并行任务和用户输入工具。
- BabyElfAGI - BabyDeerAGI和OG BabyAGI的最新修改版,具有技能类、动态任务列表和beta反射代理。
- BabyCommandAGI - 一个结合CLI和LLM的AI代理,执行shell命令和自动化任务,如应用开发和环境设置。
- BabyFoxAGI - BabyElfAGI的修改版,具有自我改进的任务列表和带并行任务的新颖聊天UI。
- BambooAI - 一个为非程序员设计的半自主AI代理,用于数据探索和分析。
- BeeBot - 一个早期研究项目,专注于开发一个用于实用任务的自主AI助手。
- Bloop - 一个由GPT-4驱动的语义代码搜索引擎,具有AI辅助的精确代码导航功能。
- BondAI - 一个多功能的自主AI代理,具有CLI、API和Docker镜像,整合了各种工具和应用程序。
订会议和总结日程。
26. CAMEL - 一个用于研究自主和交流代理的平台,包括AI用户代理和助手代理。
27. ChatDev - 一个虚拟软件公司模拟,代理承担软件开发中的不同角色。
28. ChemCrow - 一个用于合成、药物发现和材料设计任务的化学代理,将专家设计的工具与LLM结合。
29. Clippy - 一个编码任务助手,能够自主规划、编写、调试和测试项目。
30. Cody by ajhous44 - 一个AI助手,可使用自然语言交互查询代码库,提高代码导航效率。
31. Cody by Sourcegraph - 一个代码助手,通过阅读整个代码库和代码图来编写代码和回答问题。
32. Continue - 一个开源软件开发自动驾驶系统,将ChatGPT与VS Code整合,用于各种编码任务。
33. Cursor - 一个以AI为先的代码编辑器,设计用于与AI进行双人编程,提高软件开发效率。
34. Databerry - 一个无代码平台,用于创建基于用户提供数据训练的AI聊天机器人,提供各种集成和功能。
35. Deepnote AI Copilot: 一个AI代码副驾驶,为数据探索笔记本提供高效且上下文相关的代码建议。
36. DemoGPT: 利用提示快速生成基于LLM的应用程序演示。
37. DevGPT: 一个虚拟开发团队,基于AI创建微服务。
38. DevOpsGPT: 一个AI驱动的软件开发自动化解决方案。
39. dotagent: 一个代理管理系统,用于强大的AI应用和实验性自主代理。
40. English Compiler: 基于AI的编译器,用于将基于英语的markdown规格转换为功能性代码。
41. Friday: 一个开发助手,用于构建带有无限提示的Node.js项目。
42. GeniA: 一个为工程团队设计的AI平台,可在开发和云环境中执行任务。
43. Godmode: 受Auto-GPT和BabyAGI启发,通过用户界面执行各种任务。
44. GPT Engineer: 一个基于提示生成整个代码库的AI代理。
45. GPT Migrate: 促进框架或语言之间代码库的迁移。
46. GPT Pilot: 在开发人员监督实施的同时编写可扩展应用。
47. GPT Researcher: 生成任何主题的详细、事实和无偏见的研究报告。
48. GPT Runner: 管理和运行AI预设,允许与选定文件进行对话。
49. IX: 一个用于构建、调试和部署协作代理和认知工作流的平台。
50. JARVIS: 将LLM与机器学习社区连接起来,用于任务规划和执行。
- Langroid - 一个用于构建基于大型语言模型(LLM)的应用程序的Python框架。它直观、轻量且可扩展,支持多代理、多LLM系统。它包括专门的代理,如
DocChatAgent
、SQLChatAgent
和TableChatAgent
。
- Lemon Agent - 一个用于工作流自动化的计划-验证-解决(PVS)代理。它包括一个规划代理和一个解决代理,用于在GitHub、HubSpot或Airtable等各种工具上进行准确可靠的操作。
- LLM Agents - 一个用于构建利用大型语言模型自动化任务的极简库,通过命令循环和工具集成来实现。
- Local GPT - 一项允许在设备上本地与文档对话的计划,确保100%的隐私,因为没有数据离开设备。
- Loop GPT - 作为一个模块化且可扩展的Python包,重新实现Auto-GPT,专注于最小化提示开销和人类参与的反馈。
- LocalGPT - 受privateGPT启发,该项目允许使用个人文档作为信息来源,在本地设备上运行,使用Vicuna-7B模型和InstructorEmbeddings。
- MemGPT - 一个管理LLM中不同内存层的系统,提供超出LLM有限上下文窗口的扩展上下文,使得能够与数据聊天并创建持久的聊天机器人。
- Mentat - 一个用于编码协助的AI工具,协调多个位置和文件的编辑,与Copilot和ChatGPT不同,具有对项目的上下文理解。
- MetaGPT - 一个多代理框架,可以从一行需求出发,输出各种软件开发方面的内容。它为GPT分配不同的角色,形成一个协作的软件实体。
- Mini AGI - 一个基于GPT-3.5 / GPT-4的最小通用自主代理,能够执行各种任务,如分析股价和订购比萨。
- Multiagent Debate - 一项关于通过多代理辩论改善LLM中的事实性和推理的论文实现。
- Multi GPT - 一个实验性的多代理系统,其中多个“expertGPT”协作执行任务,每个代理都有自己的记忆和沟通能力。
- OpenAgents - 一个开放的语言代理平台,为研究人员、开发人员和普通用户提供以用户为中心、完全开源且可扩展的环境。
- OpenAGI - 一个AGI研发平台,使代理能够执行基准任务和开放式任务,支持各种语言模型和多模态工具学习。
- Open Interpreter - 一个开源解释器,允许LLM在计算机上运行代码以完成任务,在本地工作以确保安全的文档交互。
- Pezzo - 一个开发工具包,简化了提示设计、版本管理、发布、协作、故障排除和AI提示的可观察性。
- Private GPT - 一个用于与文档进行私密交互的工具,离线运行,并使用LangChain、GPT4All、LLaMA、Chroma和SentenceTransformers构建。
- PromethAI - 一个个性化的AI助手,协助营养和其他目标,分解问题并提供带有自动流程的解决方案。
- React Agent - 一个开源的React.js自主LLM代理,设计用于根据用户故事生成和组合React组件。
- Smol developer - 一个专门从事整体程序合成、调试和反编译的AI初级开发人员,利用GPT-4进行代码生成。
- Superagent - 一个用于创建无需编码的AI代理的工具,简化了LLM代理的配置和生产部署。
- SuperAGI - 一个开源框架,用于开发和部署自主AI代理,具有代理轨迹微调和性能遥测等功能。
- Suspicion Agent - 一个应用GPT-4学到的知识来玩具有心智理论意识的不完美信息游戏的代理。
- Sweep - 一个GitHub助手,帮助修复小bug和实现小功能,自动化代码开发和调试任务。
- Taxy AI - 一个完整的浏览器自动化工具,使用GPT-4控制浏览器并代表用户执行重复性操作。
- Teenage AGI - 一个受BabyAGI启发的AI代理,具有无限记忆回溯、深思熟虑的行动规划和记忆持久性。
- Westworld simulation - 一个多代理模拟库,用于模拟和优化具有多个交互代理的系统,受Unity软件启发。
- Voyager - 一个由LLM驱动的自主代理,在Minecraft中持续探索和技能获取,无需人类干预。
- WorkGPT - 一个用于调用API的GPT代理框架,旨在通过与AI的一系列对话来自动完成任务,直到指令完成。
- XAgent - 一个由大型语言模型(LLM)驱动的实验性自主代理,能够自主解决各种任务
- yAgents - 一个代理构建代理,用于设计、编码和调试AI驱动的工具和代理,对任何技术背景的用户都易于访问。
- Yourgoal - BabyAGI的Swift实现,一个使用OpenAI和Pinecone APIs的AI驱动任务管理系统。
分类
- AI辅助开发与代码生成:
- Includes: DevGPT, DevOpsGPT, Friday, GPT Engineer, GPT Migrate, GPT Pilot, English Compiler, Smol developer, Sweep, Taxy AI, WorkGPT, yAgents, Yourgoal.
- AI研究、分析与教育工具:
- Includes: GPT Researcher, Deepnote AI Copilot, DemoGPT, Bloop (Code Search), ChemCrow (Chemistry).
- 基于代理的系统、框架与管理工具:
- Includes: Agent4Rec, Agents, AgentVerse, AI Legion, BabyAGI, BabyBeeAGI, BabyCatAGI, BabyDeerAGI, BabyElfAGI, BabyFoxAGI, dotagent, IX, JARVIS, Superagent, SuperAGI, Teenage AGI, XAgent.
- AI应用、平台与特定任务解决方案:
- Includes: AgentGPT, AutoGen, Autonomous HR Chatbot, AutoPR, GeniA, Godmode, GPT Runner, BambooAI, BeeBot, BondAI, Cal.ai, CAMEL, ChatDev, Databerry, Clippy, Cody by ajhous44, Cody by Sourcegraph, Continue, Cursor.
- 高级AI应用、模拟与实验项目:
- Includes: Suspicion Agent, Westworld simulation, Voyager, MetaGPT, Mini AGI.
- AI框架、库与隐私保护计算:
- Includes: Langroid, LLM Agents, LocalGPT, Loop GPT, MemGPT, Mini AGI, Multi GPT, OpenAGI, Private GPT.
- 协作、多代理与智能系统:
- Includes: Multiagent Debate, MetaGPT, Lemon Agent, OpenAgents, Pezzo, PromethAI, React Agent.
- AI增强的开发与协作工具:
- Includes: Mentat, Open Interpreter, Collaborative and Multi-Agent Systems (Multiagent Debate, Multi GPT, MetaGPT), AI-Powered Tools and Applications (Pezzo, PromethAI, React Agent)
精选了3个
1 - On Generative Agents in Recommendation
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2F279f3f6f-4727-4e54-81b1-3b66a423b320%2F12f3c45a-eea5-4b86-b999-40620f1cced1%2FUntitled.png?table=block&id=d732d1ce-48f3-4597-9976-87d614a0cf15)
Agent4Rec,一个推荐系统模拟器,拥有 1,000 个 LLM 授权的生成代理。这些代理是从 MovieLens-1M 数据集初始化的,体现了不同的社会特征和偏好。每个智能体以逐页的方式与个性化电影推荐进行交互,并进行观看、评分、评价、退出、采访等各种动作。通过 Agent4Rec,我们希望探索 LLM 授权的生成代理在模拟推荐环境中真实、独立的人类行为方面的潜力。
核心功能和特点
- 生成性代理(Generative Agents):该模拟器集成了1,000个由大型语言模型(LLM)赋能的生成性代理。这些代理基于MovieLens-1M数据集初始化,模拟具有多样化社会特征和偏好的用户。
- 用户交互模拟:代理通过逐页的方式与个性化电影推荐互动,并执行各种行为,如观看、评分、评估、退出和采访,以模拟真实用户在推荐环境中的行为。
- 不同推荐系统的支持:Agent4Rec支持多种推荐系统,包括随机推荐、热门项目推荐、矩阵分解、MultVAE、LightGCN等,并允许用户通过命令行参数调整模拟配置。
实验环境与设置
- 安装依赖项:
- 首先需要设置虚拟环境并手动安装PyTorch。
- 然后,通过运行
pip install -r requirements.txt
安装所有在requirements.txt
文件中列出的依赖项。 - 这些实验在Python 3.9.12和PyTorch 1.13.1+cu117环境下测试通过。
- 设置必要环境:
- 需要在含有
setup.py
的recommenders/
目录下运行python setup.py build_ext --inplace
命令。 - 这个命令会安装加速推荐评估所需的工具。
- 运行模拟:
- 在主目录下(可以找到
main.py
),首先需要导出OpenAI API密钥:export OPENAI_API_KEY=<Your OpenAI API key>
,将<Your OpenAI API key>
替换为您自己的密钥。 - 通过运行
python main.py
命令,可以启动一个包含3个代理的基本模拟。
- 高级设置:
- 可以通过不同的命令行参数来探索各种推荐设置,例如:
python main.py --simulation_name MyExp --modeltype MF --n_avatars 10 --max_pages 5 --items_per_page 4 --execution_mode parallel
。 - 这个命令会启动一个名为MyExp的模拟,其中包含10个代理,每个代理浏览最多5页,每页4个项目。
结果查看与成本
- 查看结果:
- 模拟的结果会保存在
storage/ml-1m/<model_name>/<experiment_name>
目录中。 - 例如,如果运行上述的“探索各种推荐设置”部分中的命令,结果将保存在
storage/ml-1m/MyExp
目录中。 - 代理0的所有互动历史记录在
storage/ml-1m/MF/MyExp/running_logs/0.txt
中。
- 成本计算:
- 实验成本由ChatGPT-3.5驱动。
- 完整的涉及1000个用户的模拟大约需要16美元,即每用户大约0.016美元。
- 这意味着如果你模拟的用户数量增减,总成本将相应变化,可以通过乘以用户数来简单估算。
2 - AgentVerse
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2F279f3f6f-4727-4e54-81b1-3b66a423b320%2F3ac9654b-8910-4e22-8799-f61fb349d510%2FUntitled.png?table=block&id=db299703-823f-4f02-a4ed-f77edfb95f92)
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2F279f3f6f-4727-4e54-81b1-3b66a423b320%2Fa9a5041d-10c0-4545-92f0-56e8db46529c%2FUntitled.png?table=block&id=336a2841-9dc8-43f1-ad96-f8f31add07fe)
AgentVerse 旨在促进在各种应用程序中部署多个基于 LLM 的代理。 AgentVerse主要提供两个框架:任务解决和模拟。
任务解决:该框架将多个智能体组装成一个自动多智能体系统(AgentVerse-Tasksolving,Multi-agent as system)来协同完成相应的任务。应用领域:软件开发系统、咨询系统等。
模拟:该框架允许用户设置自定义环境来观察多个代理之间的行为或与多个代理交互。 ⚠️⚠️⚠️ 我们正在重构代码。如果您需要专门支持模拟框架的稳定版本,可以使用release-0.1分支。应用:游戏、基于LLM的代理的社会行为研究等。
特点和功能
- 多智能体协作与模拟:AgentVerse允许用户运行和测试多智能体环境,其中包括模拟应用如NLP教室和囚徒困境。这些环境模拟现实世界场景,使智能体可以在这些环境中相互作用和学习。
- 任务解决框架:该框架支持智能体在特定任务中的决策和学习。它提供了一种机制,使智能体能够解决复杂问题,并通过与环境的交互来提高它们的表现。
- 兼容和扩展性:AgentVerse支持与OpenAI API和Azure OpenAI服务的集成,允许用户使用多种语言模型,包括本地模型如LLaMA和Vicunna。
- 实用性和教育应用:除了研究和开发用途外,AgentVerse还提供了适合教育和娱乐的应用,例如简单的游戏和互动式模拟。
实验环境设置
手动安装(推荐)
确保你的Python版本至少为3.9。按照以下步骤进行安装:
如果你想使用本地模型如LLaMA,还需要安装额外的依赖:
通过pip安装
你也可以通过pip进行安装:
设置环境变量
设置OpenAI API密钥和Azure OpenAI服务(如果需要):
场景1 - 模拟环境
AgentVerse支持多智能体模拟环境。
所需模块
agentverse
agents
simulation_agent
environments
simulation_env
创建和运行多智能体环境,例如NLP教室:
图形界面 - 运行本地网站演示:
访问 http://127.0.0.1:7860/ 查看课堂环境。
场景2 - 任务解决
所需模块
agentverse
agents
simulation_env
environments
tasksolving_env
运行基准数据集实验:
运行特定问题实验:
本地使用
安装额外的依赖
如果使用本地模型,需要安装额外的依赖:
启动本地服务器
修改模型路径和名称,启动本地服务器:
修改配置文件
在配置文件中设置
llm_type
为local
,并指定model
名称:3 - Communicative Agents for Software Development
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2F279f3f6f-4727-4e54-81b1-3b66a423b320%2F3e213b7c-46f8-4862-8852-788b6b6f0814%2FUntitled.png?table=block&id=be1cd6d0-f719-4a71-a82e-69f0b7e9f4be)
用自然语言制作客制化软件
ChatDev是一家虚拟软件公司,通过不同角色的智能代理运作,包括首席执行官、首席产品官、首席技术官、程序员、审查员、测试人员和美术设计师。这些代理形成了一个多代理组织结构,并以“通过编程革新数字世界”为使命而团结在一起。 ChatDev内部的代理通过参加专业功能研讨会进行协作,包括设计、编码、测试和文档编写等任务。
ChatDev的主要目标是提供易于使用,高度可定制和可扩展的框架,该框架基于大型语言模型(LLMs),并作为研究集体智能的理想场景。
近期推出了多个新功能
- 作为软件即服务(SaaS)平台发布,方便软件开发者和企业家高效、低成本地构建软件。
- 引入增量开发功能,允许代理在现有代码上进行开发。
- 支持Docker安全执行和Git版本控制。
- 推出人机交互模式和艺术模式,使用户能够直接参与软件开发过程。
安装
克隆GitHub仓库
设置Python环境
确保您的系统中安装了Python 3.9或更高版本。使用以下命令创建并激活一个新的conda环境(您可以将
ChatDev_conda_env
替换为您选择的环境名)。安装依赖项
设置密钥
对于Unix/Linux系统
对于Windows系统
构建
使用以下命令来构建您的软件,记得替换
[description_of_your_idea]
和[project_name]
为您的项目描述和名称。对于Unix/Linux系统
对于Windows系统
运行
一旦生成,可以在
WareHouse
目录下的特定项目文件夹中找到您的软件,如project_name_DefaultOrganization_timestamp
。使用以下命令在该目录中运行。对于Unix/Linux系统
对于Windows系统
Last update: 2023-11-24
icon
password
I am currently in Melbourne
I am Looking for a Data Analytics Role…