[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-plexe-ai--plexe":3,"tool-plexe-ai--plexe":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":79,"owner_twitter":79,"owner_website":80,"owner_url":81,"languages":82,"stars":98,"forks":99,"last_commit_at":100,"license":101,"difficulty_score":23,"env_os":102,"env_gpu":103,"env_ram":103,"env_deps":104,"category_tags":117,"github_topics":118,"view_count":23,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":127,"updated_at":128,"faqs":129,"releases":158},1976,"plexe-ai\u002Fplexe","plexe","✨ Build a machine learning model from a prompt","plexe让非专业开发者也能轻松创建机器学习模型。只需用自然语言描述任务（如“预测房价”），提供数据集，系统自动完成数据处理、模型训练和部署包生成。传统建模需大量编码和调参，而plexe通过AI代理自动处理，大幅降低门槛。适合开发者、数据科学家及业务人员快速验证想法。其多代理架构在6个阶段中协作，支持XGBoost、LightGBM等主流框架，生成的模型包独立于plexe，可直接部署到任何环境。Docker镜像预装所有依赖，开箱即用。","\u003Cdiv align=\"center\">\n\n# plexe ✨\n\n[![PyPI version](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fplexe.svg)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fplexe\u002F)\n[![Discord](https:\u002F\u002Fimg.shields.io\u002Fdiscord\u002F1300920499886358529?logo=discord&logoColor=white)](https:\u002F\u002Fdiscord.gg\u002FSefZDepGMv)\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fplexe-ai_plexe_readme_6ed65bef79ec.png\" alt=\"backed-by-yc\" width=\"20%\">\n\n\nBuild machine learning models using natural language.\n\n[Quickstart](#1-quickstart) |\n[Features](#2-features) |\n[Installation](#3-installation) |\n[Documentation](#4-documentation)\n\n\u003Cbr>\n\n**plexe** lets you create machine learning models by describing them in plain language. Simply explain what you want,\nprovide a dataset, and the AI-powered system builds a fully functional model through an automated agentic approach.\nAlso available as a [managed cloud service](https:\u002F\u002Fplexe.ai).\n\n\u003Cbr>\n\nWatch the demo on YouTube:\n[![Building an ML model with Plexe](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fplexe-ai_plexe_readme_875b84f7e93c.png)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=bUwCSglhcXY)\n\u003C\u002Fdiv>\n\n## 1. Quickstart\n\n### Installation\n```bash\npip install plexe\nexport OPENAI_API_KEY=\u003Cyour-key>\nexport ANTHROPIC_API_KEY=\u003Cyour-key>\n```\n\n### Using plexe\n\nProvide a tabular dataset (Parquet, CSV, ORC, or Avro) and a natural language intent:\n\n```bash\npython -m plexe.main \\\n    --train-dataset-uri data.parquet \\\n    --intent \"predict whether a passenger was transported\" \\\n    --max-iterations 5\n```\n\n```python\nfrom plexe.main import main\nfrom pathlib import Path\n\nbest_solution, metrics, report = main(\n    intent=\"predict whether a passenger was transported\",\n    data_refs=[\"train.parquet\"],\n    max_iterations=5,\n    work_dir=Path(\".\u002Fworkdir\"),\n)\nprint(f\"Performance: {best_solution.performance:.4f}\")\n```\n\n## 2. Features\n\n### 2.1. 🤖 Multi-Agent Architecture\nThe system uses 14 specialized AI agents across a 6-phase workflow to:\n- Analyze your data and identify the ML task\n- Select the right evaluation metric\n- Search for the best model through hypothesis-driven iteration\n- Evaluate model performance and robustness\n- Package the model for deployment\n\n### 2.2. 🎯 Automated Model Building\nBuild complete models with a single call. Plexe supports **XGBoost**, **CatBoost**, **LightGBM**, **Keras**, and **PyTorch** for tabular data:\n\n```python\nbest_solution, metrics, report = main(\n    intent=\"predict house prices based on property features\",\n    data_refs=[\"housing.parquet\"],\n    max_iterations=10,                    # Search iterations\n    allowed_model_types=[\"xgboost\"],      # Or let plexe choose\n    enable_final_evaluation=True,         # Evaluate on held-out test set\n)\n```\n\nRun `python -m plexe.main --help` for all CLI options.\n\nThe output is a self-contained model package at `work_dir\u002Fmodel\u002F` (also archived as `model.tar.gz`).\nThe package has no dependency on `plexe` — build the model with plexe, deploy it anywhere:\n\n```\nmodel\u002F\n├── artifacts\u002F          # Trained model + feature pipeline (pickle)\n├── src\u002F                # Inference predictor, pipeline code, training template\n├── schemas\u002F            # Input\u002Foutput JSON schemas\n├── config\u002F             # Hyperparameters\n├── evaluation\u002F         # Metrics and detailed analysis reports\n├── model.yaml          # Model metadata\n└── README.md           # Usage instructions with example code\n```\n\n### 2.3. 🐳 Batteries-Included Docker Images\nRun plexe with everything pre-configured — PySpark, Java, and all dependencies included.\nA `Makefile` is provided for common workflows:\n\n```bash\nmake build          # Build the Docker image\nmake test-quick     # Fast sanity check (~1 iteration)\nmake run-titanic    # Run on Spaceship Titanic dataset\n```\n\nOr run directly:\n\n```bash\ndocker run --rm \\\n    -e OPENAI_API_KEY=$OPENAI_API_KEY \\\n    -e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY \\\n    -v $(pwd)\u002Fdata:\u002Fdata -v $(pwd)\u002Fworkdir:\u002Fworkdir \\\n    plexe:py3.12 python -m plexe.main \\\n        --train-dataset-uri \u002Fdata\u002Fdataset.parquet \\\n        --intent \"predict customer churn\" \\\n        --work-dir \u002Fworkdir \\\n        --spark-mode local\n```\n\nA `config.yaml` in the project root is automatically mounted. A Databricks Connect image\nis also available: `docker build --target databricks .`\n\n### 2.4. ⚙️ YAML Configuration\nCustomize LLM routing, search parameters, Spark settings, and more via a config file:\n\n```yaml\n# config.yaml\nmax_search_iterations: 5\nallowed_model_types: [xgboost, catboost]\nspark_driver_memory: \"4g\"\nhypothesiser_llm: \"openai\u002Fgpt-5-mini\"\nfeature_processor_llm: \"anthropic\u002Fclaude-sonnet-4-5-20250929\"\n```\n\n```bash\nCONFIG_FILE=config.yaml python -m plexe.main ...\n```\n\nSee [`config.yaml.template`](config.yaml.template) for all available options.\n\n### 2.5. 🌐 Multi-Provider LLM Support\nPlexe uses LLMs via [LiteLLM](https:\u002F\u002Fdocs.litellm.ai\u002Fdocs\u002Fproviders), so you can use any supported provider:\n\n```yaml\n# Route different agents to different providers\nhypothesiser_llm: \"openai\u002Fgpt-5-mini\"\nfeature_processor_llm: \"anthropic\u002Fclaude-sonnet-4-5-20250929\"\nmodel_definer_llm: \"ollama\u002Fllama3\"\n```\n\n> [!NOTE]\n> Plexe *should* work with most LiteLLM providers, but we actively test only with `openai\u002F*` and `anthropic\u002F*`\n> models. If you encounter issues with other providers, please let us know.\n\n### 2.6. 📊 Experiment Dashboard\nVisualize experiment results, search trees, and evaluation reports with the built-in Streamlit dashboard:\n\n```bash\npython -m plexe.viz --work-dir .\u002Fworkdir\n```\n\n### 2.7. 🔌 Extensibility\nConnect plexe to custom storage, tracking, and deployment infrastructure via the `WorkflowIntegration` interface:\n\n```python\nmain(intent=\"...\", data_refs=[...], integration=MyCustomIntegration())\n```\n\nSee [`plexe\u002Fintegrations\u002Fbase.py`](plexe\u002Fintegrations\u002Fbase.py) for the full interface.\n\n## 3. Installation\n\n### 3.1. Installation Options\n```bash\npip install plexe                    # Core (XGBoost, Keras, scikit-learn)\n```\n\nYou can add optional dependencies either by framework or by task grouping:\n- Framework extras: `catboost`, `lightgbm`, `pytorch`\n- Task extras: `tabular` (CatBoost + LightGBM), `vision` (PyTorch)\n- Platform extras: `pyspark`, `aws`\n\nExamples:\n```bash\npip install \"plexe[tabular,pyspark]\"   # tabular stack + local PySpark\npip install \"plexe[pytorch,aws]\"       # explicit framework + S3 support\n```\n\nRequires Python >= 3.10, \u003C 3.13.\n\n### 3.2. API Keys\n```bash\nexport OPENAI_API_KEY=\u003Cyour-key>\nexport ANTHROPIC_API_KEY=\u003Cyour-key>\n```\nSee [LiteLLM providers](https:\u002F\u002Fdocs.litellm.ai\u002Fdocs\u002Fproviders) for all supported providers.\n\n## 4. Documentation\nFor full documentation, visit [docs.plexe.ai](https:\u002F\u002Fdocs.plexe.ai).\n\n## 5. Contributing\nSee [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines. Join our [Discord](https:\u002F\u002Fdiscord.gg\u002FSefZDepGMv) to connect with the team.\n\n## 6. License\n[Apache-2.0 License](LICENSE)\n\n## 7. Citation\nIf you use Plexe in your research, please cite it as follows:\n\n```bibtex\n@software{plexe2025,\n  author = {De Bernardi, Marcello AND Dubey, Vaibhav},\n  title = {Plexe: Build machine learning models using natural language.},\n  year = {2025},\n  publisher = {GitHub},\n  howpublished = {\\url{https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe}},\n}\n```\n","\u003Cdiv align=\"center\">\n\n# plexe ✨\n\n[![PyPI version](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fplexe.svg)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fplexe\u002F)\n[![Discord](https:\u002F\u002Fimg.shields.io\u002Fdiscord\u002F1300920499886358529?logo=discord&logoColor=white)](https:\u002F\u002Fdiscord.gg\u002FSefZDepGMv)\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fplexe-ai_plexe_readme_6ed65bef79ec.png\" alt=\"backed-by-yc\" width=\"20%\">\n\n\n用自然语言构建机器学习模型。\n\n[快速入门](#1-quickstart) |\n[功能](#2-features) |\n[安装](#3-installation) |\n[文档](#4-documentation)\n\n\u003Cbr>\n\n**plexe** 让你只需用通俗易懂的语言描述，就能创建机器学习模型。只需简单说明你的需求，提供一个数据集，由人工智能驱动的系统便会通过自动化代理方式构建出一个功能完备的模型。也可作为[托管云服务](https:\u002F\u002Fplexe.ai)使用。\n\n\u003Cbr>\n\n观看 YouTube 上的演示：\n[![用 Plexe 构建 ML 模型](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fplexe-ai_plexe_readme_875b84f7e93c.png)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=bUwCSglhcXY)\n\u003C\u002Fdiv>\n\n## 1. 快速入门\n\n### 安装\n```bash\npip install plexe\nexport OPENAI_API_KEY=\u003Cyour-key>\nexport ANTHROPIC_API_KEY=\u003Cyour-key>\n```\n\n### 使用 plexe\n\n提供一个表格数据集（Parquet、CSV、ORC 或 Avro）和一个自然语言意图：\n\n```bash\npython -m plexe.main \\\n    --train-dataset-uri data.parquet \\\n    --intent \"预测乘客是否被运送\" \\\n    --max-iterations 5\n```\n\n```python\nfrom plexe.main import main\nfrom pathlib import Path\n\nbest_solution, metrics, report = main(\n    intent=\"预测乘客是否被运送\",\n    data_refs=[\"train.parquet\"],\n    max_iterations=5,\n    work_dir=Path(\".\u002Fworkdir\"),\n)\nprint(f\"性能：{best_solution.performance:.4f}\")\n```\n\n## 2. 功能\n\n### 2.1. 🤖 多智能体架构\n该系统在 6 阶段的工作流中使用了 14 个专业 AI 智能体，以：\n- 分析你的数据并识别 ML 任务\n- 选择合适的评估指标\n- 通过假设驱动的迭代搜索最佳模型\n- 评估模型性能与鲁棒性\n- 打包模型以便部署\n\n### 2.2. 🎯 自动化模型构建\n只需一次调用即可构建完整的模型。Plexe 支持针对表格数据的 **XGBoost**、**CatBoost**、**LightGBM**、**Keras** 和 **PyTorch**：\n\n```python\nbest_solution, metrics, report = main(\n    intent=\"根据房产特征预测房价\",\n    data_refs=[\"housing.parquet\"],\n    max_iterations=10,                    # 搜索迭代次数\n    allowed_model_types=[\"xgboost\"],      # 或让 Plexe 自行选择\n    enable_final_evaluation=True,         # 在保留的测试集上评估\n)\n```\n\n运行 `python -m plexe.main --help` 查看所有 CLI 选项。\n\n输出是一个自包含的模型包，位于 `work_dir\u002Fmodel\u002F`（也存档为 `model.tar.gz`）。该包不依赖于 `plexe`——用 Plexe 构建模型后，可随处部署：\n\n```\nmodel\u002F\n├── artifacts\u002F          # 训练好的模型 + 特征流水线 (pickle)\n├── src\u002F                # 推理预测器、流水线代码、训练模板\n├── schemas\u002F            # 输入\u002F输出 JSON 模式\n├── config\u002F             # 超参数\n├── evaluation\u002F         # 指标与详细分析报告\n├── model.yaml          # 模型元数据\n└── README.md           # 使用说明与示例代码\n```\n\n### 2.3. 🐳 附带完整依赖的 Docker 镜像\n运行 Plexe 时已预配置好一切——包括 PySpark、Java 和所有依赖项。提供了一个 `Makefile` 用于常见工作流：\n\n```bash\nmake build          # 构建 Docker 镜像\nmake test-quick     # 快速健康检查 (~1 次迭代)\nmake run-titanic    # 在 Spaceship Titanic 数据集上运行\n```\n\n或直接运行：\n\n```bash\ndocker run --rm \\\n    -e OPENAI_API_KEY=$OPENAI_API_KEY \\\n    -e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY \\\n    -v $(pwd)\u002Fdata:\u002Fdata -v $(pwd)\u002Fworkdir:\u002Fworkdir \\\n    plexe:py3.12 python -m plexe.main \\\n        --train-dataset-uri \u002Fdata\u002Fdataset.parquet \\\n        --intent \"预测客户流失\" \\\n        --work-dir \u002Fworkdir \\\n        --spark-mode local\n```\n\n项目根目录中的 `config.yaml` 会自动挂载。还提供了一个 Databricks Connect 镜像：`docker build --target databricks .`\n\n### 2.4. ⚙️ YAML 配置\n通过配置文件自定义 LLM 路由、搜索参数、Spark 设置等：\n\n```yaml\n# config.yaml\nmax_search_iterations: 5\nallowed_model_types: [xgboost, catboost]\nspark_driver_memory: \"4g\"\nhypothesiser_llm: \"openai\u002Fgpt-5-mini\"\nfeature_processor_llm: \"anthropic\u002Fclaude-sonnet-4-5-20250929\"\n```\n\n```bash\nCONFIG_FILE=config.yaml python -m plexe.main ...\n```\n\n查看 [`config.yaml.template`](config.yaml.template) 了解所有可用选项。\n\n### 2.5. 🌐 多提供商 LLM 支持\nPlexe 通过 [LiteLLM](https:\u002F\u002Fdocs.litellm.ai\u002Fdocs\u002Fproviders) 使用 LLM，因此你可以使用任何支持的提供商：\n\n```yaml\n# 将不同智能体路由到不同提供商\nhypothesiser_llm: \"openai\u002Fgpt-5-mini\"\nfeature_processor_llm: \"anthropic\u002Fclaude-sonnet-4-5-20250929\"\nmodel_definer_llm: \"ollama\u002Fllama3\"\n```\n\n> [!NOTE]\n> Plexe 应该能与大多数 LiteLLM 提供商配合，但我们仅主动测试 `openai\u002F*` 和 `anthropic\u002F*` 模型。如果你遇到其他提供商的问题，请告诉我们。\n\n### 2.6. 📊 实验仪表板\n使用内置的 Streamlit 仪表板可视化实验结果、搜索树和评估报告：\n\n```bash\npython -m plexe.viz --work-dir .\u002Fworkdir\n```\n\n### 2.7. 🔌 可扩展性\n通过 `WorkflowIntegration` 接口将 Plexe 连接到自定义存储、追踪和部署基础设施：\n\n```python\nmain(intent=\"...\", data_refs=[...], integration=MyCustomIntegration())\n```\n\n查看 [`plexe\u002Fintegrations\u002Fbase.py`](plexe\u002Fintegrations\u002Fbase.py) 了解完整接口。\n\n## 3. 安装\n\n### 3.1. 安装选项\n```bash\npip install plexe                    # 核心（XGBoost、Keras、scikit-learn）\n```\n\n你可以按框架或任务分组添加可选依赖：\n- 框架额外依赖：`catboost`、`lightgbm`、`pytorch`\n- 任务额外依赖：`tabular`（CatBoost + LightGBM）、`vision`（PyTorch）\n- 平台额外依赖：`pyspark`、`aws`\n\n示例：\n```bash\npip install \"plexe[tabular,pyspark]\"   # 表格堆栈 + 本地 PySpark\npip install \"plexe[pytorch,aws]\"       # 显式框架 + S3 支持\n```\n\n需要 Python >= 3.10，\u003C 3.13。\n\n### 3.2. API 密钥\n```bash\nexport OPENAI_API_KEY=\u003Cyour-key>\nexport ANTHROPIC_API_KEY=\u003Cyour-key>\n```\n查看 [LiteLLM 提供商](https:\u002F\u002Fdocs.litellm.ai\u002Fdocs\u002Fproviders) 了解所有支持的提供商。\n\n## 4. 文档\n如需完整文档，请访问 [docs.plexe.ai](https:\u002F\u002Fdocs.plexe.ai)。\n\n## 5. 贡献\n请参阅 [CONTRIBUTING.md](CONTRIBUTING.md) 了解指南。加入我们的 [Discord](https:\u002F\u002Fdiscord.gg\u002FSefZDepGMv) 与团队交流。\n\n## 6. 许可证\n[Apache-2.0 许可证](LICENSE)\n\n## 7. 引用\n如果您在研究中使用Plexe，请按以下方式引用：\n\n```bibtex\n@software{plexe2025,\n  author = {德·贝纳尔迪，马塞洛 AND 杜贝，瓦伊巴夫},\n  title = {Plexe：利用自然语言构建机器学习模型。},\n  year = {2025},\n  publisher = {GitHub},\n  howpublished = {\\url{https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe}},\n}\n```","# 环境准备\n- Python 3.10 至 3.12\n- OpenAI 和 Anthropic API 密钥（需自行注册）\n\n# 安装步骤\n推荐使用清华镜像加速安装：\n```bash\npip install -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple plexe\n```\n\n设置环境变量：\n```bash\nexport OPENAI_API_KEY=\u003Cyour-key>\nexport ANTHROPIC_API_KEY=\u003Cyour-key>\n```\n\n# 基本使用\n使用命令行快速构建模型：\n```bash\npython -m plexe.main \\\n    --train-dataset-uri data.parquet \\\n    --intent \"预测乘客是否被运输\" \\\n    --max-iterations 5\n```\n\n或者使用 Python 代码：\n```python\nfrom plexe.main import main\nfrom pathlib import Path\n\nbest_solution, metrics, report = main(\n    intent=\"预测乘客是否被运输\",\n    data_refs=[\"train.parquet\"],\n    max_iterations=5,\n    work_dir=Path(\".\u002Fworkdir\"),\n)\nprint(f\"模型性能: {best_solution.performance:.4f}\")\n```","某中型电商公司的数据分析团队正试图构建一个客户流失预测模型，以优化营销预算分配。团队仅有两名数据工程师，缺乏专职机器学习工程师，且业务方要求两周内上线可用模型。\n\n### 没有 plexe 时\n- 数据工程师需手动清洗数据、探索特征分布、判断是分类还是回归问题，耗时近3天。\n- 团队对XGBoost、LightGBM等模型的超参数调优经验不足，尝试了5种模型后仍无法稳定达到0.82以上的AUC。\n- 模型训练完成后，需手动编写API接口、封装依赖、生成输入输出Schema，耗费额外4天开发时间。\n- 模型部署文档由工程师手写，业务人员反馈“看不懂如何用”，导致上线后频繁咨询。\n- 整个流程依赖个人经验，一旦成员请假，项目进度立即停滞。\n\n### 使用 plexe 后\n- 只需上传客户行为数据（CSV）并输入自然语言指令“预测未来30天是否会流失”，plexe在12小时内自动完成数据理解、特征工程与模型选择。\n- plexe自动测试了XGBoost、CatBoost等模型，最终输出AUC达0.87的模型，性能优于团队此前所有手动尝试。\n- 自动生成完整可部署模型包，包含推理代码、JSON Schema、训练模板和README，直接上传至生产环境无需修改。\n- 模型包附带通俗易懂的使用示例，市场团队可直接用Python一行代码调用预测接口，无需技术支援。\n- 整个流程可复用，下次构建“高价值客户识别”模型时，仅需更换数据和意图，5小时内即可产出新模型。\n\nplexe让非AI专家的团队也能在一周内交付生产级机器学习模型，真正实现了“用语言构建AI”。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fplexe-ai_plexe_b35d3673.png","plexe-ai","Plexe AI","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fplexe-ai_43de73e4.png","Production-ready specialised AI - All from natural language!",null,"https:\u002F\u002Fplexe.ai","https:\u002F\u002Fgithub.com\u002Fplexe-ai",[83,87,90,94],{"name":84,"color":85,"percentage":86},"Python","#3572A5",97.2,{"name":88,"color":89,"percentage":23},"Makefile","#427819",{"name":91,"color":92,"percentage":93},"Dockerfile","#384d54",0.6,{"name":95,"color":96,"percentage":97},"Shell","#89e051",0.3,2556,252,"2026-04-04T15:55:32","Apache-2.0","Linux, macOS, Windows","未说明",{"notes":105,"python":106,"dependencies":107},"需要配置 OpenAI 或 Anthropic API 密钥；支持通过 Docker 运行，内置 PySpark 和 Java 依赖；建议使用 config.yaml 自定义 LLM 路由和 Spark 设置；模型输出为独立包，部署无需依赖 plexe",">=3.10,\u003C3.13",[108,109,110,111,112,113,114,115,116],"xgboost","catboost","lightgbm","torch","keras","scikit-learn","pyspark","litellm","streamlit",[14,13,15],[119,120,121,122,123,124,125,126],"agentic-ai","agents","machine-learning","ml","mlengineering","mlops","multiagent","ai","2026-03-27T02:49:30.150509","2026-04-06T05:15:30.653933",[130,135,140,145,150,154],{"id":131,"question_zh":132,"answer_zh":133,"source_url":134},8912,"如何解决推理错误？","升级到 smolmodels 版本 0.12.6，使用命令 `pip install smolmodels==0.12.6`。","https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fissues\u002F83",{"id":136,"question_zh":137,"answer_zh":138,"source_url":139},8913,"使用 openLLMs 时出现 JSONDecodeError 如何解决？","确保使用支持结构化输出的模型；库已添加检查，只允许支持结构化输出的模型，否则会抛出 ValueError。详情见 #66。","https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fissues\u002F60",{"id":141,"question_zh":142,"answer_zh":143,"source_url":144},8914,"如何获取项目支持？","加入 Discord 服务器：https:\u002F\u002Fdiscord.gg\u002FSefZDepGMv。","https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fissues\u002F39",{"id":146,"question_zh":147,"answer_zh":148,"source_url":149},8915,"Discord 链接无效如何解决？","Discord 链接已修复，请使用 https:\u002F\u002Fdiscord.gg\u002FSefZDepGMv。","https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fissues\u002F48",{"id":151,"question_zh":152,"answer_zh":153,"source_url":144},8916,"如何贡献代码到项目？","查看 open issues，解决它们；或在 Discord 服务器中寻求支持：https:\u002F\u002Fdiscord.gg\u002FSefZDepGMv。",{"id":155,"question_zh":156,"answer_zh":157,"source_url":134},8917,"为什么模型训练超时？","默认超时为5分钟，升级到 smolmodels 0.12.6 版本修复此问题。",[159,164,169,174,179,184,189,194,199,204,209,214,219,224,229,234,239,244,249,254],{"id":160,"version":161,"summary_zh":162,"released_at":163},106327,"v1.4.4","## What's Changed\r\n* chore: ignore .claude and .codex directories by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F188\r\n* feat: support explicit split inputs and nn epoch overrides by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F189\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.4.3...v1.4.4","2026-03-06T11:41:06",{"id":165,"version":166,"summary_zh":167,"released_at":168},106328,"v1.4.3","## What's Changed\r\n* feat: add probability-aware metric support across search and evaluation by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F187\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.4.2...v1.4.3","2026-03-03T05:27:15",{"id":170,"version":171,"summary_zh":172,"released_at":173},106329,"v1.4.2","## What's Changed\r\n* fix: calibrate evaluator verdict rubric and baseline context by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F186\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.4.1...v1.4.2","2026-03-03T01:13:01",{"id":175,"version":176,"summary_zh":177,"released_at":178},106330,"v1.4.1","## What's Changed\r\n* feat: add train\u002Fval gap tracking and bootstrap insights by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F185\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.4.0...v1.4.1","2026-03-03T00:29:28",{"id":180,"version":181,"summary_zh":182,"released_at":183},106331,"v1.4.0","## What's Changed\r\n* feat: add gpu-aware keras\u002Fpytorch training runtime by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F183\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.3.6...v1.4.0","2026-03-02T22:20:59",{"id":185,"version":186,"summary_zh":187,"released_at":188},106332,"v1.3.5","## What's Changed\r\n* test: add staged pytest integration suite by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F182\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.3.4...v1.3.5","2026-03-02T16:55:43",{"id":190,"version":191,"summary_zh":192,"released_at":193},106333,"v1.3.4","## What's Changed\r\n* fix: enable keras and final evaluation make tests by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F177\r\n* feat: add reproducibility controls and deterministic search RNG by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F179\r\n* feat: add model card packaging by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F180\r\n* feat: add column exclusion pipeline by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F181\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.3.0...v1.3.4","2026-03-02T13:22:47",{"id":195,"version":196,"summary_zh":197,"released_at":198},106334,"v1.3.0","## What's Changed\r\n* feat: conditional framework extras by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F176\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.2.2...v1.3.0","2026-02-27T14:07:56",{"id":200,"version":201,"summary_zh":202,"released_at":203},106335,"v1.2.2","## What's Changed\r\n* fix: improve feature name resolution by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F175\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.2.1...v1.2.2","2026-02-27T11:28:33",{"id":205,"version":206,"summary_zh":207,"released_at":208},106336,"v1.2.1","## What's Changed\r\n* fix: make torch optional for packaging by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F174\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.2.0...v1.2.1","2026-02-26T15:36:43",{"id":210,"version":211,"summary_zh":212,"released_at":213},106337,"v1.2.0","## What's Changed\r\n* feat: add pytorch support by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F173\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.1.0...v1.2.0","2026-02-26T14:01:23",{"id":215,"version":216,"summary_zh":217,"released_at":218},106338,"v1.1.0","## What's Changed\r\n* fix(setup): resolve path bugs and missing config extras by @Virtuoso633 in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F164\r\n* ci: allow codex branch prefix by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F167\r\n* chore: add AGENTS instructions by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F169\r\n* test: expand unit coverage for core utilities by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F171\r\n* feat: add lightgbm support by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F172\r\n\r\n## New Contributors\r\n* @Virtuoso633 made their first contribution in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F164\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.0.1...v1.1.0","2026-02-25T16:47:03",{"id":220,"version":221,"summary_zh":222,"released_at":223},106339,"v1.0.1","## What's Changed\r\n* fix: miscellaneous rewrite errors by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F162\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv1.0.0...v1.0.1","2026-02-09T15:45:09",{"id":225,"version":226,"summary_zh":227,"released_at":228},106340,"v1.0.0","## What's Changed\r\n* fix: multi-agent mermaid chart by @ptdatta in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F138\r\n* Implemented smart auto-scroll in chat [closes #142] by @AumPatel1 in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F143\r\n* feat: Update example by @JasonHonKL in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F149\r\n* chore: update cicd workflows by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F160\r\n* feat: rewrite model builder to use guardrailed workflow by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F161\r\n\r\n## New Contributors\r\n* @ptdatta made their first contribution in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F138\r\n* @AumPatel1 made their first contribution in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F143\r\n* @JasonHonKL made their first contribution in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F149\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv0.26.2...v1.0.0","2026-02-09T11:02:53",{"id":230,"version":231,"summary_zh":232,"released_at":233},106341,"v0.26.2","## What's Changed\r\n* hotfix: lock sklearn to 1.6.1 for backwards compatibility by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F137\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv0.26.1...v0.26.2","2025-06-06T20:42:45",{"id":235,"version":236,"summary_zh":237,"released_at":238},106342,"v0.26.1","## What's Changed\r\n* hotfix: missing deprecated library by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F136\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv0.26.0...v0.26.1","2025-06-06T19:20:14",{"id":240,"version":241,"summary_zh":242,"released_at":243},106343,"v0.26.0","## What's Changed\r\n* chore: update dependencies\u002Fextras grouping by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F133\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv0.25.0...v0.26.0","2025-06-06T03:08:35",{"id":245,"version":246,"summary_zh":247,"released_at":248},106344,"v0.25.0","## What's Changed\r\n* feature: support lists in model schemas by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F132\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv0.24.0...v0.25.0","2025-05-30T19:42:44",{"id":250,"version":251,"summary_zh":252,"released_at":253},106345,"v0.24.0","## What's Changed\r\n* feat: dynamic model schema resolution by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F130\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv0.23.6...v0.24.0","2025-05-30T08:42:31",{"id":255,"version":256,"summary_zh":257,"released_at":258},106346,"v0.23.6","## What's Changed\r\n* fix: improve safety of yaml serialisation by @marcellodebernardi in https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fpull\u002F129\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fplexe-ai\u002Fplexe\u002Fcompare\u002Fv0.23.5...v0.23.6","2025-05-29T18:44:44"]