From 70a557cf4ae2f21b60d8b08435b7e9f64b53bbbd Mon Sep 17 00:00:00 2001 From: zachary62 Date: Mon, 31 Mar 2025 22:00:15 -0400 Subject: [PATCH] init batch tutorial --- cookbook/pocketflow-batch/README.md | 79 ++++++++++++ cookbook/pocketflow-batch/main.py | 68 ++++++++++ cookbook/pocketflow-batch/requirements.txt | 3 + .../translations/README_CHINESE.md | 120 ++++++++++++++++++ cookbook/pocketflow-batch/utils.py | 20 +++ .../pocketflow-majority-vote/requirements.txt | 3 +- 6 files changed, 292 insertions(+), 1 deletion(-) create mode 100644 cookbook/pocketflow-batch/README.md create mode 100644 cookbook/pocketflow-batch/main.py create mode 100644 cookbook/pocketflow-batch/requirements.txt create mode 100644 cookbook/pocketflow-batch/translations/README_CHINESE.md create mode 100644 cookbook/pocketflow-batch/utils.py diff --git a/cookbook/pocketflow-batch/README.md b/cookbook/pocketflow-batch/README.md new file mode 100644 index 0000000..dc142b3 --- /dev/null +++ b/cookbook/pocketflow-batch/README.md @@ -0,0 +1,79 @@ +# Majority Vote Reasoning + +This project demonstrates a majority vote implementation that enables LLMs to solve complex reasoning problems by aggregating multiple independent attempts. It's designed to improve problem-solving accuracy through consensus-based reasoning. + +## Features + +- Improves model reliability on complex problems through multiple attempts +- Works with models like Claude 3.7 Sonnet +- Solves problems that single attempts often fail on +- Provides detailed reasoning traces for verification +- Uses a consensus approach to reduce the impact of occasional reasoning errors + +## Getting Started + +1. Install the required packages: +```bash +pip install -r requirements.txt +``` + +2. Set up your API key: +```bash +export ANTHROPIC_API_KEY="your-api-key-here" +``` + +3. Run a test problem to see majority voting in action: +```bash +python main.py +``` + +4. Try your own reasoning problem: +```bash +python main.py --problem "Your complex reasoning problem here" --tries 5 +``` + +## How It Works + +The implementation uses a MajorityVoteNode that processes multiple attempts and finds consensus: + +```mermaid +flowchart LR + mv[MajorityVoteNode] +``` + +The MajorityVoteNode: +1. Makes multiple independent attempts to solve the same problem +2. Collects structured answers from each attempt +3. Determines the most frequent answer as the final solution +4. Returns the consensus answer + +This approach helps overcome occasional reasoning errors that might occur in individual attempts. + +## Example Problem + +Example Problem from [Quant Interview](https://www.youtube.com/watch?v=SCP7JptxPU0): + +``` +You work at a shoe factory. In front of you, there are three pairs of shoes (six individual shoes) with the following sizes: two size 4s, two size 5s, and two size 6s. The factory defines an "acceptable pair" as two shoes that differ in size by a maximum of one size (e.g., a size 5 and a size 6 would be an acceptable pair). If you close your eyes and randomly pick three pairs of shoes without replacement, what is the probability that you end up drawing three acceptable pairs? +``` + +Below is an example of how the majority vote approach uses Claude 3.7 Sonnet to solve this complex problem: + +``` +======================== +All structured answers: ['0.333', '0.333', '0.333', '0.6', '0.333'] +Majority vote => 0.333 +Frequency => 4 +======================== + +=== Final Answer === +0.333 +==================== +``` + +This shows that 4 out of 5 attempts yielded the same answer (0.333), which is chosen as the final solution. + +## Files + +- [`main.py`](./main.py): Implementation of the majority vote node and flow +- [`utils.py`](./utils.py): Simple wrapper for calling the Anthropic model \ No newline at end of file diff --git a/cookbook/pocketflow-batch/main.py b/cookbook/pocketflow-batch/main.py new file mode 100644 index 0000000..b8ff940 --- /dev/null +++ b/cookbook/pocketflow-batch/main.py @@ -0,0 +1,68 @@ +import os +from pocketflow import BatchNode, Flow +from utils import call_llm + +class TranslateTextNode(BatchNode): + def prep(self, shared): + text = shared.get("text", "(No text provided)") + languages = shared.get("languages", ["Chinese", "Spanish", "Japanese", "German", + "Russian", "Portuguese", "French", "Korean"]) + + # Create batches for each language translation + return [(text, lang) for lang in languages] + + def exec(self, data_tuple): + text, language = data_tuple + + prompt = f""" +Please translate the following markdown file into {language}. +But keep the original markdown format, links and code blocks. +Directly return the translated text, without any other text or comments. + +Original: +{text} + +Translated:""" + + result = call_llm(prompt) + + print(f"Translated {language} text") + + return {"language": language, "translation": result} + + def post(self, shared, prep_res, exec_res_list): + # Create output directory if it doesn't exist + output_dir = shared.get("output_dir", "translations") + os.makedirs(output_dir, exist_ok=True) + + # Write each translation to a file + for result in exec_res_list: + language, translation = result["language"], result["translation"] + + # Write to file + filename = os.path.join(output_dir, f"README_{language.upper()}.md") + with open(filename, "w", encoding="utf-8") as f: + f.write(translation) + + print(f"Saved translation to {filename}") + +if __name__ == "__main__": + # read the text from ../../README.md + with open("../../README.md", "r") as f: + text = f.read() + + # Default settings + shared = { + "text": text, + "languages": ["Chinese", "Spanish", "Japanese", "German", "Russian", "Portuguese", "French", "Korean"], + "output_dir": "translations" + } + + # Run the translation flow + translate_node = TranslateTextNode(max_retries=3) + flow = Flow(start=translate_node) + flow.run(shared) + + print("\n=== Translation Complete ===") + print(f"Translations saved to: {shared['output_dir']}") + print("============================") \ No newline at end of file diff --git a/cookbook/pocketflow-batch/requirements.txt b/cookbook/pocketflow-batch/requirements.txt new file mode 100644 index 0000000..9fe93b4 --- /dev/null +++ b/cookbook/pocketflow-batch/requirements.txt @@ -0,0 +1,3 @@ +pocketflow>=0.0.1 +anthropic>=0.15.0 +pyyaml>=6.0 \ No newline at end of file diff --git a/cookbook/pocketflow-batch/translations/README_CHINESE.md b/cookbook/pocketflow-batch/translations/README_CHINESE.md new file mode 100644 index 0000000..822e779 --- /dev/null +++ b/cookbook/pocketflow-batch/translations/README_CHINESE.md @@ -0,0 +1,120 @@ +
+ +
+ + +![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg) +[![Docs](https://img.shields.io/badge/docs-latest-blue)](https://the-pocket.github.io/PocketFlow/) + + + + +Pocket Flow 是一个仅[100行代码](https://github.com/The-Pocket/PocketFlow/blob/main/pocketflow/__init__.py)的极简主义 LLM 框架 + +- **轻量级**：仅100行代码。零臃肿，零依赖，零供应商锁定。 + +- **表达力强**：拥有你喜爱的一切——([多](https://the-pocket.github.io/PocketFlow/design_pattern/multi_agent.html))[智能体](https://the-pocket.github.io/PocketFlow/design_pattern/agent.html)、[工作流](https://the-pocket.github.io/PocketFlow/design_pattern/workflow.html)、[RAG](https://the-pocket.github.io/PocketFlow/design_pattern/rag.html)等等。 + +- **[智能体编程](https://zacharyhuang.substack.com/p/agentic-coding-the-most-fun-way-to)**：让AI智能体（如Cursor AI）构建智能体—生产力提升10倍！ + +- 安装：```pip install pocketflow```或直接复制[源代码](https://github.com/The-Pocket/PocketFlow/blob/main/pocketflow/__init__.py)（仅100行）。 + +- 了解更多，请查看[文档](https://the-pocket.github.io/PocketFlow/)。了解动机，请阅读[故事](https://zacharyhuang.substack.com/p/i-built-an-llm-framework-in-just)。 + +- 🎉 加入我们的[Discord](https://discord.gg/hUHHE9Sa6T)！ + +- 🎉 感谢[@zvictor](https://www.github.com/zvictor)、[@jackylee941130](https://www.github.com/jackylee941130)和[@ZebraRoy](https://www.github.com/ZebraRoy)，我们现在有了[TypeScript版本](https://github.com/The-Pocket/PocketFlow-Typescript)！ + +## 为什么选择Pocket Flow？ + +当前的LLM框架过于臃肿...LLM框架只需要100行代码！ + +
+ + + + | | **抽象** | **应用特定包装器** | **供应商特定包装器** | **代码行数** | **大小** | +|----------------|:-----------------------------: |:-----------------------------------------------------------:|:------------------------------------------------------------:|:---------------:|:----------------------------:| +| LangChain | 智能体, 链 | 很多
(例如，问答，摘要) | 很多
(例如，OpenAI, Pinecone等) | 405K | +166MB | +| CrewAI | 智能体, 链 | 很多
(例如，FileReadTool, SerperDevTool) | 很多
(例如，OpenAI, Anthropic, Pinecone等) | 18K | +173MB | +| SmolAgent | 智能体 | 一些
(例如，CodeAgent, VisitWebTool) | 一些
(例如，DuckDuckGo, Hugging Face等) | 8K | +198MB | +| LangGraph | 智能体, 图 | 一些
(例如，语义搜索) | 一些
(例如，PostgresStore, SqliteSaver等) | 37K | +51MB | +| AutoGen | 智能体 | 一些
(例如，工具智能体，聊天智能体) | 很多 [可选]
(例如，OpenAI, Pinecone等) | 7K
(仅核心) | +26MB
(仅核心) | +| **PocketFlow** | **图** | **无** | **无** | **100** | **+56KB** | + +
+ +## Pocket Flow如何工作？ + +这[100行代码](https://github.com/The-Pocket/PocketFlow/blob/main/pocketflow/__init__.py)捕捉了LLM框架的核心抽象：图！ +
+
+ +
+
+ +从此出发，很容易实现流行的设计模式，如([多](https://the-pocket.github.io/PocketFlow/design_pattern/multi_agent.html))[智能体](https://the-pocket.github.io/PocketFlow/design_pattern/agent.html)、[工作流](https://the-pocket.github.io/PocketFlow/design_pattern/workflow.html)、[RAG](https://the-pocket.github.io/PocketFlow/design_pattern/rag.html)等。 +
+
+ +
+
+✨ 以下是基础教程： + +
+ +| 名称 | 难度 | 描述 | +| :-------------: | :-------------: | :--------------------- | +| [聊天](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-chat) | ☆☆☆
*极简* | 具有对话历史的基础聊天机器人 | +| [结构化输出](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-structured-output) | ☆☆☆
*极简* | 通过提示从简历中提取结构化数据 | +| [工作流](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-workflow) | ☆☆☆
*极简* | 一个包含大纲、内容写作和风格应用的写作工作流 | +| [智能体](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-agent) | ☆☆☆
*极简* | 一个能够搜索网络并回答问题的研究智能体 | +| [RAG](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-rag) | ☆☆☆
*极简* | 一个简单的检索增强生成过程 | +| [映射-归约](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-map-reduce) | ☆☆☆
*极简* | 使用映射-归约模式进行批量评估的简历资格处理器 | +| [流式处理](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-llm-streaming) | ☆☆☆
*极简* | 具有用户中断功能的实时LLM流式演示 | +| [聊天护栏](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-chat-guardrail) | ☆☆☆
*极简* | 仅处理与旅行相关查询的旅行顾问聊天机器人 | +| [多智能体](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-multi-agent) | ★☆☆
*初级* | 一个用于两个智能体之间异步通信的禁忌词游戏 | +| [监督者](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-supervisor) | ★☆☆
*初级* | 研究智能体变得不可靠...让我们构建一个监督流程 | +| [并行](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-parallel-batch) | ★☆☆
*初级* | 展示3倍加速的并行执行演示 | +| [并行流](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-parallel-batch-flow) | ★☆☆
*初级* | 使用多个过滤器展示8倍加速的并行图像处理演示 | +| [多数投票](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-majority-vote) | ★☆☆
*初级* | 通过聚合多个解决方案尝试提高推理准确性 | +| [思考](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-thinking) | ★☆☆
*初级* | 通过思维链解决复杂推理问题 | +| [记忆](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-chat-memory) | ★☆☆
*初级* | 具有短期和长期记忆的聊天机器人 | +| [MCP](https://github.com/The-Pocket/PocketFlow/tree/main/cookbook/pocketflow-mcp) | ★☆☆
*初级* | 使用模型上下文协议进行数值运算的智能体 | + +
+ +👀 想看更多极简教程？[创建一个issue！](https://github.com/The-Pocket/PocketFlow/issues/new) + +## 如何使用Pocket Flow？ + +🚀 通过**智能体编程**——最快的LLM应用开发范式，其中*人类设计*而*智能体编码*！ + +
+
+ + IMAGE ALT TEXT + +
+
+ +✨ 以下是更复杂LLM应用的示例： + +
+ +| 应用名称 | 难度 | 主题 | 人类设计 | 智能体代码 | +| :-------------: | :-------------: | :---------------------: | :---: | :---: | +| [用Cursor构建Cursor](https://github.com/The-Pocket/Tutorial-Cursor)
我们很快就会达到奇点... | ★★★
*高级* | [智能体](https://the-pocket.github.io/PocketFlow/design_pattern/agent.html) | [设计文档](https://github.com/The-Pocket/Tutorial-Cursor/blob/main/docs/design.md) | [流程代码](https://github.com/The-Pocket/Tutorial-Cursor/blob/main/flow.py) +| [咨询AI Paul Graham](https://github.com/The-Pocket/Tutorial-YC-Partner)
咨询AI版Paul Graham，以防你未被录取 | ★★☆
*中级* | [RAG](https://the-pocket.github.io/PocketFlow/design_pattern/rag.html)
[映射-归约](https://the-pocket.github.io/PocketFlow/design_pattern/mapreduce.html)
[TTS](https://the-pocket.github.io/PocketFlow/utility_function/text_to_speech.html) | [设计文档](https://github.com/The-Pocket/Tutorial-AI-Paul-Graham/blob/main/docs/design.md) | [流程代码](https://github.com/The-Pocket/Tutorial-AI-Paul-Graham/blob/main/flow.py) +| [Youtube摘要器](https://github.com/The-Pocket/Tutorial-Youtube-Made-Simple)
像对五岁孩子一样向你解释YouTube视频 | ★☆☆
*初级* | [映射-归约](https://the-pocket.github.io/PocketFlow/design_pattern/mapreduce.html) | [设计文档](https://github.com/The-Pocket/Tutorial-Youtube-Made-Simple/blob/main/docs/design.md) | [流程代码](https://github.com/The-Pocket/Tutorial-Youtube-Made-Simple/blob/main/flow.py) +| [冷启动生成器](https://github.com/The-Pocket/Tutorial-Cold-Email-Personalization)
即时破冰器，让冷门线索变热门 | ★☆☆
*初级* | [映射-归约](https://the-pocket.github.io/PocketFlow/design_pattern/mapreduce.html)
[网络搜索](https://the-pocket.github.io/PocketFlow/utility_function/websearch.html) | [设计文档](https://github.com/The-Pocket/Tutorial-Cold-Email-Personalization/blob/master/docs/design.md) | [流程代码](https://github.com/The-Pocket/Tutorial-Cold-Email-Personalization/blob/master/flow.py) + +
+ +- 想学习**智能体编程**？ + + - 查看[我的YouTube](https://www.youtube.com/@ZacharyLLM?sub_confirmation=1)，了解上述应用如何制作的视频教程！ + + - 想构建自己的LLM应用？阅读这篇[文章](https://zacharyhuang.substack.com/p/agentic-coding-the-most-fun-way-to)！从[这个模板](https://github.com/The-Pocket/PocketFlow-Template-Python)开始！ + + - 想了解详细步骤？阅读这份[指南](https://the-pocket.github.io/PocketFlow/guide.html)！ \ No newline at end of file diff --git a/cookbook/pocketflow-batch/utils.py b/cookbook/pocketflow-batch/utils.py new file mode 100644 index 0000000..afa1817 --- /dev/null +++ b/cookbook/pocketflow-batch/utils.py @@ -0,0 +1,20 @@ +from anthropic import Anthropic +import os + +def call_llm(prompt): + client = Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY", "your-api-key")) + response = client.messages.create( + model="claude-3-7-sonnet-20250219", + max_tokens=10000, + messages=[ + {"role": "user", "content": prompt} + ] + ) + return response.content[0].text + +if __name__ == "__main__": + print("## Testing call_llm") + prompt = "In a few words, what is the meaning of life?" + print(f"## Prompt: {prompt}") + response = call_llm(prompt) + print(f"## Response: {response}") \ No newline at end of file diff --git a/cookbook/pocketflow-majority-vote/requirements.txt b/cookbook/pocketflow-majority-vote/requirements.txt index bf3b8c2..9fe93b4 100644 --- a/cookbook/pocketflow-majority-vote/requirements.txt +++ b/cookbook/pocketflow-majority-vote/requirements.txt @@ -1,2 +1,3 @@ pocketflow>=0.0.1 -anthropic>=0.15.0 # For Claude API access \ No newline at end of file +anthropic>=0.15.0 +pyyaml>=6.0 \ No newline at end of file