[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-build-with-groq--g1":3,"tool-build-with-groq--g1":61},[4,18,26,36,44,53],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":10,"last_commit_at":24,"category_tags":25,"status":17},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,"2026-04-05T11:01:52",[14,15,13],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"status":17},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",155373,2,"2026-04-14T11:34:08",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"status":17},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。",108322,"2026-04-10T11:39:34",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":10,"last_commit_at":59,"category_tags":60,"status":17},4487,"LLMs-from-scratch","rasbt\u002FLLMs-from-scratch","LLMs-from-scratch 是一个基于 PyTorch 的开源教育项目，旨在引导用户从零开始一步步构建一个类似 ChatGPT 的大型语言模型（LLM）。它不仅是同名技术著作的官方代码库，更提供了一套完整的实践方案，涵盖模型开发、预训练及微调的全过程。\n\n该项目主要解决了大模型领域“黑盒化”的学习痛点。许多开发者虽能调用现成模型，却难以深入理解其内部架构与训练机制。通过亲手编写每一行核心代码，用户能够透彻掌握 Transformer 架构、注意力机制等关键原理，从而真正理解大模型是如何“思考”的。此外，项目还包含了加载大型预训练权重进行微调的代码，帮助用户将理论知识延伸至实际应用。\n\nLLMs-from-scratch 特别适合希望深入底层原理的 AI 开发者、研究人员以及计算机专业的学生。对于不满足于仅使用 API，而是渴望探究模型构建细节的技术人员而言，这是极佳的学习资源。其独特的技术亮点在于“循序渐进”的教学设计：将复杂的系统工程拆解为清晰的步骤，配合详细的图表与示例，让构建一个虽小但功能完备的大模型变得触手可及。无论你是想夯实理论基础，还是为未来研发更大规模的模型做准备",90106,"2026-04-06T11:19:32",[35,15,13,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":66,"readme_en":67,"readme_zh":68,"quickstart_zh":69,"use_case_zh":70,"hero_image_url":71,"owner_login":72,"owner_name":73,"owner_avatar_url":74,"owner_bio":75,"owner_company":76,"owner_location":76,"owner_email":77,"owner_twitter":78,"owner_website":79,"owner_url":80,"languages":81,"stars":86,"forks":87,"last_commit_at":88,"license":89,"difficulty_score":32,"env_os":90,"env_gpu":91,"env_ram":92,"env_deps":93,"category_tags":101,"github_topics":102,"view_count":32,"oss_zip_url":76,"oss_zip_packed_at":76,"status":17,"created_at":104,"updated_at":105,"faqs":106,"releases":107},7549,"build-with-groq\u002Fg1","g1","g1: Using Llama-3.1 70b on Groq to create o1-like reasoning chains","g1 是一个开源实验项目，旨在通过巧妙的提示词策略，让现有的开源大模型（如 Llama-3.1 70B）具备类似 OpenAI o1 的深层推理能力。它主要解决了传统大模型在面对逻辑陷阱或复杂数学问题时容易“直觉犯错”的痛点，例如经典的\"strawberry 中有几个 r\"问题，g1 能显著提升此类任务的准确率。\n\n该工具的核心亮点在于不依赖昂贵的额外训练，而是利用 Groq 的高速推理服务，引导模型生成动态的“思维链”。在回答问题时，g1 会强制模型分步骤思考，主动探索多种解题路径、质疑初步结论并自我纠错，最终将完整的推理过程以可视化的形式呈现给用户。这种机制让模型从“直接猜测”转变为“严谨推导”。\n\ng1 非常适合 AI 开发者、研究人员以及对大模型推理机制感兴趣的技术爱好者使用。开发者可以基于其开源代码探索更先进的提示工程策略；研究人员可将其作为研究思维链（Chain of Thought）有效性的实验床；而普通技术用户也能通过其提供的界面，直观体验开源模型在处理逻辑难题时的进化潜力。作为一个早期原型，g1 展示了仅凭提示词优化即可释放开源模型巨大潜能的可行性。","# g1: Using Llama-3.1 70b on Groq to create o1-like reasoning chains\n\n[Video Demo](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fdb2a221f-f8eb-48c3-b5a7-8399c6300243)\n\nThis is an early prototype of using prompting strategies to improve the LLM's reasoning capabilities through o1-like reasoning chains. This allows the LLM to \"think\" and solve logical problems that usually otherwise stump leading models. Unlike o1, all the reasoning tokens are shown, and the app uses an open source model.\n\ng1 is experimental and being open sourced to help inspire the open source community to develop new strategies to produce o1-like reasoning. This experiment helps show the power of prompting reasoning in visualized steps, not a comparison to or full replication of o1, which uses different techniques. OpenAI's o1 is instead trained with large-scale reinforcement learning to reason using Chain of Thought, achieving state-of-the-art performance on complex PhD-level problems. \n\ng1 demonstrates the potential of prompting alone to overcome straightforward LLM logic issues like the Strawberry problem, allowing existing open source models to benefit from dynamic reasoning chains and an improved interface for exploring them.\n\n\n### How it works\n\ng1 powered by Llama3.1-70b creates reasoning chains, in principle a dynamic Chain of Thought, that allows the LLM to \"think\" and solve some logical problems that usually otherwise stump leading models.\n\nAt each step, the LLM can choose to continue to another reasoning step, or provide a final answer. Each step is titled and visible to the user. The system prompt also includes tips for the LLM. There is a full explanation under Prompt Breakdown, but a few examples are asking the model to “include exploration of alternative answers” and “use at least 3 methods to derive the answer”.\n\nThe reasoning ability of the LLM is therefore improved through combining Chain-of-Thought with the requirement to try multiple methods, explore alternative answers, question previous draft solutions, and consider the LLM’s limitations. This alone, without any training, is sufficient to achieve ~70% accuracy on the Strawberry problem (n=10, \"How many Rs are in strawberry?\"). Without prompting, Llama-3.1-70b had 0% accuracy and ChatGPT-4o had 30% accuracy.\n\n\n### Examples\n\n> [!IMPORTANT]\n> g1 is not perfect, but it can perform significantly better than LLMs out-of-the-box. From initial testing, g1 accurately solves simple logic problems 60-80% of the time that usually stump LLMs. However, accuracy has yet to be formally evaluated. See examples below.\n\n\n##### How many Rs are in strawberry?\n\nPrompt: How many Rs are in strawberry?\n\nResult:\n\n![Strawberry example](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fbuild-with-groq_g1_readme_059ffc3ebcc5.png)\n\n---\n\nPrompt: Which is larger, .9 or .11?\n\nResult:\n\n![0.9 or 0.11 example](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fbuild-with-groq_g1_readme_cf52b01bcd10.png)\n\n\n### Quickstart\n\nTo use the Streamlit UI, follow these instructions:\n\n~~~\npython3 -m venv venv\n~~~\n\n~~~\nsource venv\u002Fbin\u002Factivate\n~~~\n\n~~~\npip3 install -r requirements.txt\n~~~\n\n~~~\nexport GROQ_API_KEY=gsk...\n~~~\n\n~~~\nstreamlit run app.py\n~~~\n\n---\n\nAlternatively, follow these additional instructions to use the Gradio UI:\n\n~~~\ncd gradio\n~~~\n\n~~~\npip3 install -r requirements.txt\n~~~\n\n~~~\npython3 app.py\n~~~\n\n\n### Prompting Strategy\n\nThe prompt is as follows:\n\n```\nYou are an expert AI assistant that explains your reasoning step by step. For each step, provide a title that describes what you're doing in that step, along with the content. Decide if you need another step or if you're ready to give the final answer. Respond in JSON format with 'title', 'content', and 'next_action' (either 'continue' or 'final_answer') keys. USE AS MANY REASONING STEPS AS POSSIBLE. AT LEAST 3. BE AWARE OF YOUR LIMITATIONS AS AN LLM AND WHAT YOU CAN AND CANNOT DO. IN YOUR REASONING, INCLUDE EXPLORATION OF ALTERNATIVE ANSWERS. CONSIDER YOU MAY BE WRONG, AND IF YOU ARE WRONG IN YOUR REASONING, WHERE IT WOULD BE. FULLY TEST ALL OTHER POSSIBILITIES. YOU CAN BE WRONG. WHEN YOU SAY YOU ARE RE-EXAMINING, ACTUALLY RE-EXAMINE, AND USE ANOTHER APPROACH TO DO SO. DO NOT JUST SAY YOU ARE RE-EXAMINING. USE AT LEAST 3 METHODS TO DERIVE THE ANSWER. USE BEST PRACTICES.\n\nExample of a valid JSON response:\njson\n{\n    \"title\": \"Identifying Key Information\",\n    \"content\": \"To begin solving this problem, we need to carefully examine the given information and identify the crucial elements that will guide our solution process. This involves...\",\n    \"next_action\": \"continue\"\n}\n```\n\n#### Breakdown\n\nFirst, a persona is added:\n\n> You are an expert AI assistant that explains your reasoning step by step.\n\n\n\nThen, instructions to describe the expected step-by-step reasoning process while titling each reasoning step. This includes the ability for the LLM to decide if another reasoning step is needed or if the final answer can be provided.\n\n> For each step, provide a title that describes what you're doing in that step, along with the content. Decide if you need another step or if you're ready to give the final answer. \n\n\n\nJSON formatting is introduced with an example provided later.\n\n> Respond in JSON format with 'title', 'content', and 'next_action' (either 'continue' or 'final_answer') keys. \n\n\n\nIn all-caps to improve prompt compliance by emphasizing the importance of the instruction, a set of tips and best practices are included.\n\n1. Use as many reasoning steps as possible. At least 3. -> This ensures the LLM actually takes the time to think first, and results usually in about 5-10 steps.\n2. Be aware of your limitations as an llm and what you can and cannot do. -> This helps the LLM remember to use techniques which produce better results, like breaking \"strawberry\" down into individual letters before counting.\n3. Include exploration of alternative answers. Consider you may be wrong, and if you are wrong in your reasoning, where it would be. -> A large part of the gains seem to come from the LLM re-evaluating its initial response to ensure it logically aligns with the problem.\n4. When you say you are re-examining, actually re-examine, and use another approach to do so. Do not just say you are re-examining. -> This encourages the prevention of the LLM just saying it re-examined a problem without actually trying a new approach. \n5. Use at least 3 methods to derive the answer. -> This helps the LLM come to the right answer by trying multiple methods to derive it.\n6. Use best practices. -> This is as simple as the \"Do better\" prompts which improve LLM code output. By telling the LLM to use best practices, or do better, it generally performs better!\n\n\n> USE AS MANY REASONING STEPS AS POSSIBLE. AT LEAST 3. BE AWARE OF YOUR LIMITATIONS AS AN LLM AND WHAT YOU CAN AND CANNOT DO. IN YOUR REASONING, INCLUDE EXPLORATION OF ALTERNATIVE ANSWERS. CONSIDER YOU MAY BE WRONG, AND IF YOU ARE WRONG IN YOUR REASONING, WHERE IT WOULD BE. FULLY TEST ALL OTHER POSSIBILITIES. YOU CAN BE WRONG. WHEN YOU SAY YOU ARE RE-EXAMINING, ACTUALLY RE-EXAMINE, AND USE ANOTHER APPROACH TO DO SO. DO NOT JUST SAY YOU ARE RE-EXAMINING. USE AT LEAST 3 METHODS TO DERIVE THE ANSWER. USE BEST PRACTICES.\n\n\n\nFinally, after the problem is added as a user message, an assistant message is loaded to provide a standardized starting point for the LLM's generation.\n\n> Assistant: Thank you! I will now think step by step following my instructions, starting at the beginning after decomposing the problem\n\n\n### Top Forks\n\n* Huggingface Spaces Demo: [![Hugging Face Spaces](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fxylin\u002Fg1-demo)\n* Mult1: Using multiple AI providers to create o1-like reasoning chains ([GitHub Repository](https:\u002F\u002Fgithub.com\u002Ftcsenpai\u002Fmulti1))\n* thinkR: o1 like chain of thoughts with local LLMs in R ([GitHub Repository](https:\u002F\u002Fgithub.com\u002Feonurk\u002FthinkR))\n\n### Credits\n\nThis app was developed by Benjamin Klieger.\n","# g1：在 Groq 上使用 Llama-3.1 70B 构建类似 o1 的推理链\n\n[视频演示](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fdb2a221f-f8eb-48c3-b5a7-8399c6300243)\n\n这是一个早期原型，旨在通过提示策略构建类似 o1 的推理链，从而提升大语言模型的推理能力。这种方法使模型能够“思考”，并解决通常会让主流模型感到棘手的逻辑问题。与 o1 不同的是，g1 会展示所有的推理步骤，并且使用的是开源模型。\n\ng1 属于实验性项目，现已开源，目的是激发开源社区开发新的策略，以实现类似 o1 的推理能力。这项实验旨在展示分步可视化提示的强大作用，而非对 o1 的比较或完全复现；o1 采用了不同的技术路线。OpenAI 的 o1 是通过大规模强化学习训练而成，利用思维链进行推理，在复杂的博士级问题上达到了业界领先水平。\n\ng1 证明了仅靠提示就能克服诸如“草莓问题”这类简单的语言模型逻辑缺陷，使现有的开源模型也能受益于动态推理链以及更友好的交互界面来探索这些推理过程。\n\n\n### 工作原理\n\n由 Llama3.1-70B 驱动的 g1 能够生成推理链——本质上是一种动态的思维链——让大语言模型具备“思考”的能力，从而解决一些通常会让主流模型难以应对的逻辑问题。\n\n在每一步中，模型可以选择继续进行下一步推理，或者直接给出最终答案。每个步骤都有标题，并对用户可见。系统提示中还包含给模型的建议。完整说明请参见“提示拆解”部分，其中的一些示例包括要求模型“探索备选答案”以及“至少使用三种方法推导出答案”。\n\n通过将思维链与多方法尝试、备选答案探索、对先前草稿解决方案的质疑以及对模型自身局限性的考量相结合，g1 显著提升了大语言模型的推理能力。仅凭这一点，无需任何额外训练，便能在“草莓问题”（n=10，“草莓中有多少个 R？”）上达到约 70% 的准确率。而在未使用提示的情况下，Llama-3.1-70B 的准确率为 0%，而 ChatGPT-4o 的准确率为 30%。\n\n\n### 示例\n\n> [!重要]\n> g1 并不完美，但其表现明显优于开箱即用的大语言模型。初步测试显示，g1 在通常会让大语言模型犯难的简单逻辑问题上，准确率可达 60%–80%。不过，目前尚未对准确率进行正式评估。以下是一些示例。\n\n\n##### 草莓中有多少个 R？\n\n提示：草莓中有多少个 R？\n\n结果：\n\n![草莓示例](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fbuild-with-groq_g1_readme_059ffc3ebcc5.png)\n\n---\n\n提示：0.9 和 0.11 哪个更大？\n\n结果：\n\n![0.9 或 0.11 示例](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fbuild-with-groq_g1_readme_cf52b01bcd10.png)\n\n\n### 快速入门\n\n要使用 Streamlit 界面，请按照以下步骤操作：\n\n~~~\npython3 -m venv venv\n~~~\n\n~~~\nsource venv\u002Fbin\u002Factivate\n~~~\n\n~~~\npip3 install -r requirements.txt\n~~~\n\n~~~\nexport GROQ_API_KEY=gsk...\n~~~\n\n~~~\nstreamlit run app.py\n~~~\n\n---\n\n或者，您也可以按照以下步骤使用 Gradio 界面：\n\n~~~\ncd gradio\n~~~\n\n~~~\npip3 install -r requirements.txt\n~~~\n\n~~~\npython3 app.py\n~~~\n\n### 提示策略\n\n提示内容如下：\n\n```\n你是一位专家级的AI助手，会逐步解释你的推理过程。对于每一步，提供一个描述该步骤内容的标题，并附上具体内容。判断是否还需要进一步的步骤，或者是否可以给出最终答案。请以JSON格式回复，包含‘title’（标题）、‘content’（内容）和‘next_action’（下一步行动，可选‘continue’或‘final_answer’）三个键。尽可能多地使用推理步骤，至少3步。要意识到作为大语言模型的局限性，清楚自己能做什么、不能做什么。在推理过程中，应探索其他可能的答案，考虑自己可能会出错，并分析如果推理有误，错在哪里。充分检验所有其他可能性。你有可能犯错。当你提到正在重新审视时，务必真正地重新审视，并尝试另一种方法来实现；不要只是说说而已。至少使用3种方法来得出答案。遵循最佳实践。\n\n有效JSON响应示例：\njson\n{\n    \"title\": \"识别关键信息\",\n    \"content\": \"为了解决这个问题，我们需要仔细检查给定的信息，找出指导我们解题的关键要素。这包括...\",\n    \"next_action\": \"continue\"\n}\n```\n\n#### 分解\n\n首先，添加了一个角色设定：\n\n> 你是一位专家级的AI助手，会逐步解释你的推理过程。\n\n\n\n接着，给出了关于预期逐步推理过程的指示，要求为每个推理步骤命名，并说明其内容。这还包括让大语言模型能够决定是否需要继续推理，还是可以直接给出最终答案。\n\n> 对于每一步，提供一个描述该步骤内容的标题，并附上具体内容。判断是否还需要进一步的步骤，或者是否可以给出最终答案。\n\n\n\n随后引入了JSON格式，并提供了示例。\n\n> 请以JSON格式回复，包含‘title’（标题）、‘content’（内容）和‘next_action’（下一步行动，可选‘continue’或‘final_answer’）三个键。\n\n\n\n为了提高提示的执行效果，用大写字母强调了一组建议和最佳实践：\n\n1. 尽可能多地使用推理步骤，至少3步。——这样可以确保大语言模型真正花时间思考，通常会产生5到10个步骤。\n2. 意识到作为大语言模型的局限性，明确自己能做什么、不能做什么。——这有助于模型记住使用更有效的技巧，比如在计数之前先将“strawberry”拆分成单个字母。\n3. 探索其他可能的答案。考虑自己可能会出错，并分析如果推理有误，错在哪里。——大部分收益似乎来自于模型重新评估其初始回答，以确保其逻辑与问题一致。\n4. 当你说正在重新审视时，务必真正地重新审视，并采用另一种方法；不要只是口头说说而已。——这可以防止模型仅仅声称已经重新审视过问题，而实际上并未尝试新的方法。\n5. 至少使用3种方法来得出答案。——通过尝试多种方法，可以帮助模型找到正确的答案。\n6. 遵循最佳实践。——例如，“做得更好”这类提示可以提升大语言模型生成代码的质量。通过告诉模型要遵循最佳实践或做得更好，它通常会有更好的表现！\n\n\n> 尽可能多地使用推理步骤，至少3步。要意识到作为大语言模型的局限性，清楚自己能做什么、不能做什么。在推理过程中，应探索其他可能的答案，考虑自己可能会出错，并分析如果推理有误，错在哪里。充分检验所有其他可能性。你有可能犯错。当你提到正在重新审视时，务必真正地重新审视，并尝试另一种方法；不要只是说说而已。至少使用3种方法来得出答案。遵循最佳实践。\n\n\n\n最后，在用户消息中加入具体问题后，加载了一条助手消息，为大语言模型的生成提供一个标准化的起点。\n\n> 助手：谢谢！我现在将按照我的指示一步步思考，从分解问题开始。\n\n\n### 相关项目\n\n* Huggingface Spaces演示：[![Hugging Face Spaces](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fxylin\u002Fg1-demo)\n* Mult1：利用多个AI提供商构建类似o1的推理链（[GitHub仓库](https:\u002F\u002Fgithub.com\u002Ftcsenpai\u002Fmulti1)）\n* thinkR：在R语言中使用本地大语言模型构建类似o1的思维链（[GitHub仓库](https:\u002F\u002Fgithub.com\u002Feonurk\u002FthinkR)）\n\n### 致谢\n\n本应用由Benjamin Klieger开发。","# g1 快速上手指南\n\ng1 是一个实验性开源项目，旨在通过特定的提示词策略（Prompting Strategies），利用 Llama-3.1 70b 模型在 Groq 平台上生成类似 OpenAI o1 的推理链。它能让大模型在输出最终答案前进行多步骤“思考”，显著提升解决逻辑难题（如\"strawberry 中有几个 r\"）的能力。\n\n## 环境准备\n\n在开始之前，请确保您的开发环境满足以下要求：\n\n*   **操作系统**：Linux, macOS 或 Windows (WSL)\n*   **Python 版本**：Python 3.8 或更高版本\n*   **API Key**：您需要一个有效的 [Groq API Key](https:\u002F\u002Fconsole.groq.com\u002F) (gsk_...)\n*   **网络环境**：由于需要连接 Groq API 和下载 Python 包，请确保网络畅通。如有条件，可配置国内 pip 镜像源加速安装。\n\n## 安装步骤\n\n本项目提供两种用户界面（UI）：**Streamlit**（推荐）和 **Gradio**。以下是基于 Streamlit 的安装流程。\n\n### 1. 创建并激活虚拟环境\n\n```bash\npython3 -m venv venv\nsource venv\u002Fbin\u002Factivate\n```\n*(Windows 用户使用 `venv\\Scripts\\activate`)*\n\n### 2. 安装依赖\n\n建议使用国内镜像源（如清华源）以加快安装速度：\n\n```bash\npip3 install -r requirements.txt -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n### 3. 配置 API Key\n\n将您的 Groq API Key 导出为环境变量：\n\n```bash\nexport GROQ_API_KEY=gsk_your_actual_api_key_here\n```\n*(Windows PowerShell 用户使用 `$env:GROQ_API_KEY=\"gsk_...\"`)*\n\n## 基本使用\n\n### 启动应用\n\n运行以下命令启动 Streamlit 界面：\n\n```bash\nstreamlit run app.py\n```\n\n浏览器会自动打开本地服务地址（通常为 `http:\u002F\u002Flocalhost:8501`）。\n\n### 使用示例\n\n在界面中输入逻辑问题，观察模型的逐步推理过程。\n\n**示例问题 1：字母计数陷阱**\n> **Prompt:** How many Rs are in strawberry?\n>\n> **预期行为:** 模型不会直接回答，而是会分解单词、逐个检查字母、尝试多种计数方法并进行自我验证，最终给出正确答案（3个）。\n\n**示例问题 2：小数比较**\n> **Prompt:** Which is larger, .9 or .11?\n>\n> **预期行为:** 模型会通过数学转换或多方法对比，避免被数字长度误导，正确判断 0.9 更大。\n\n### 核心机制说明\n\ng1 的核心在于其系统提示词（System Prompt），它强制模型执行以下操作：\n1.  **分步思考**：至少进行 3 个推理步骤。\n2.  **多方法验证**：使用至少 3 种不同的方法推导答案。\n3.  **自我反思**：主动探索替代答案，质疑之前的草稿，并承认潜在的局限性。\n4.  **结构化输出**：每一步都包含标题、内容和下一步动作（继续或给出最终答案）。\n\n---\n*注：如需使用 Gradio 界面，可进入 `gradio` 目录，安装该目录下的 `requirements.txt` 后运行 `python3 app.py`。*","某数据科学团队在开发自动化报表系统时，需要让 AI 处理包含复杂逻辑陷阱和易错数学比较的原始业务数据。\n\n### 没有 g1 时\n- **陷入直觉陷阱**：面对“统计单词中某字母数量”或\"0.9 与 0.11 谁更大”这类反直觉问题，模型常因依赖概率预测而直接给出错误答案。\n- **思维过程黑盒**：模型直接输出结论，开发人员无法知晓其推理路径，难以定位是计算失误还是逻辑理解偏差。\n- **缺乏自我纠错**：模型一旦生成初步错误思路，往往不会主动质疑或尝试其他方法，导致错误结果被固化。\n- **调试成本高昂**：为了修正逻辑漏洞，工程师不得不编写大量硬编码规则或进行繁琐的人工复核，效率极低。\n\n### 使用 g1 后\n- **多路径深度推理**：g1 强制模型至少采用三种不同方法推导答案，并主动探索替代方案，有效破解了“草莓中有几个 r\"等经典逻辑陷阱。\n- **思维链可视化**：每一步推理都带有标题和内容展示，团队成员可清晰看到模型如何从“初步假设”到“自我质疑”再到“最终验证”的全过程。\n- **动态自我修正**：模型在推理过程中会主动评估自身局限性，发现前序步骤潜在错误时自动重新审视，显著提升了复杂逻辑题的准确率。\n- **零训练成本升级**：无需微调模型或收集额外训练数据，仅通过优化的提示词策略就让开源的 Llama-3.1-70b 具备了媲美顶尖闭源模型的逻辑能力。\n\ng1 通过结构化的动态思维链，将开源大模型从“直觉回答者”转变为具备自我反思能力的“逻辑推理专家”，大幅降低了复杂任务的处理门槛。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fbuild-with-groq_g1_059ffc3e.png","build-with-groq","Build With Groq","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fbuild-with-groq_aa5c3e72.png","Fully open-sourced end-to-end applications and solutions for your biggest use cases built with Groq API that you can own. Just clone, customize, and deploy.",null,"devrel@groq.com","GroqInc","https:\u002F\u002Fconsole.groq.com","https:\u002F\u002Fgithub.com\u002Fbuild-with-groq",[82],{"name":83,"color":84,"percentage":85},"Python","#3572A5",100,4195,360,"2026-04-13T11:36:50","MIT","Linux, macOS, Windows","不需要本地 GPU（模型通过 Groq API 云端运行）","未说明（取决于本地 Python 环境及浏览器负载，通常 4GB+ 即可）",{"notes":94,"python":95,"dependencies":96},"该工具不本地运行大模型，而是调用 Groq 云端的 Llama-3.1-70b 模型，因此无需高性能显卡或大量显存。使用前必须设置 GROQ_API_KEY 环境变量。主要界面依赖 Streamlit 或 Gradio，具体依赖库版本需参考项目中的 requirements.txt 文件。","3.x (README 中使用 python3 命令)",[97,98,99,100],"streamlit","gradio","requests","python-dotenv",[35,13],[103],"managed-by-terraform","2026-03-27T02:49:30.150509","2026-04-15T06:51:59.660398",[],[]]