From 70478bfc61e1f987b8dfb3aa601a7a87a620fdeb Mon Sep 17 00:00:00 2001 From: BigDong Date: Wed, 17 Jan 2024 19:37:51 +0800 Subject: [PATCH] [Docs] Update Agent docs (#590) Co-authored-by: BIGWangYuDong Co-authored-by: ZwwWayne --- agent/README.md | 22 +++++++++++ agent/README_zh-CN.md | 22 +++++++++++ agent/lagent.md | 73 ++++++++++++++++++++++++++++++++++++ agent/lagent_zh-CN.md | 73 ++++++++++++++++++++++++++++++++++++ agent/pal_inference.md | 62 ++++++++++++++++++++++++++++++ agent/pal_inference.py | 0 agent/pal_inference_zh-CN.md | 62 ++++++++++++++++++++++++++++++ 7 files changed, 314 insertions(+) create mode 100644 agent/README_zh-CN.md create mode 100644 agent/lagent.md create mode 100644 agent/lagent_zh-CN.md create mode 100644 agent/pal_inference.md create mode 100644 agent/pal_inference.py create mode 100644 agent/pal_inference_zh-CN.md diff --git a/agent/README.md b/agent/README.md index e69de29..370415b 100644 --- a/agent/README.md +++ b/agent/README.md @@ -0,0 +1,22 @@ +# InternLM-Chat Agent + +English | [简体中文](README_zh-CN.md) + +## Introduction + +InternLM-Chat-7B v1.1 has been released as the first open-source model with code interpreter capabilities, supportting external tools such as Python code interpreter and search engine. + +InternLM2-Chat, open sourced on January 17, 2024, further enhances its capabilities in code interpreter and general tool utilization. With improved and more generalized instruction understanding, tool selection, and reflection abilities, InternLM2-Chat can more reliably support complex agents and multi-step tool calling for more intricate tasks. InternLM2-Chat exhibits decent computational and reasoning abilities even without external tools, surpassing ChatGPT in mathematical performance. When combined with a code interpreter, InternLM2-Chat-20B obtains comparable results to GPT-4 on GSM8K and MATH. Leveraging strong foundational capabilities in mathematics and tools, InternLM2-Chat provides practical data analysis capabilities. + +The results of InternLM2-Chat-20B on math code interpreter is as below: + +| | GSM8K | MATH | +| :---: | :---: | :--: | +| InternLM2-Chat-20B | 79.6 | 32.5 | +| InternLM2-Chat-20B with Code Interpreter | 84.5 | 51.2 | +| ChatGPT (GPT-3.5) | 78.2 | 28.0 | +| GPT-4 | 91.4 | 45.8 | + +## Usages + +We offer examples using [Lagent](lagent.md) to build agents based on InternLM2-Chat to call code interpreter or search API. Additionally, we provide an example code using [PAL to evaluate GSM8K math problems](pal_inference.md) with InternLM-Chat-7B. diff --git a/agent/README_zh-CN.md b/agent/README_zh-CN.md new file mode 100644 index 0000000..8d1e5b2 --- /dev/null +++ b/agent/README_zh-CN.md @@ -0,0 +1,22 @@ +# InternLM-Chat 智能体 + +[English](README.md) | 简体中文 + +## 简介 + +InternLM-Chat-7B v1.1 是首个具有代码解释能力的开源对话模型，支持 Python 解释器和搜索引擎等外部工具。 + +InternLM2-Chat 进一步提高了它在代码解释和通用工具调用方面的能力。基于更强和更具有泛化性的指令理解、工具筛选与结果反思等能力，新版模型可以更可靠地支持复杂智能体的搭建，支持对工具进行有效的多轮调用，完成较复杂的任务。模型在不使用外部工具的条件下已具备不错的计算能力和推理能力，数理表现超过 ChatGPT；在配合代码解释器（code-interpreter）的条件下，InternLM2-Chat-20B 在 GSM8K 和 MATH 上可以达到和 GPT-4 相仿的水平。基于在数理和工具方面强大的基础能力，InternLM2-Chat 提供了实用的数据分析能力。 + +以下是 InternLM2-Chat-20B 在数学代码解释器上的结果。 + +| | GSM8K | MATH | +| :---: | :---: | :--: | +| InternLM2-Chat-20B 单纯依靠内在能力 | 79.6 | 32.5 | +| InternLM2-Chat-20B 配合代码解释器 | 84.5 | 51.2 | +| ChatGPT (GPT-3.5) | 78.2 | 28.0 | +| GPT-4 | 91.4 | 45.8 | + +## 体验 + +我们提供了使用 [Lagent](lagent_zh_cn.md) 来基于 InternLM2-Chat 构建智能体调用代码解释器或者搜索等工具的例子。同时，我们也提供了采用 [PAL 评测 GSM8K 数学题](pal_inference_zh-CN.md) InternLM-Chat-7B 的样例。 diff --git a/agent/lagent.md b/agent/lagent.md new file mode 100644 index 0000000..209d9b0 --- /dev/null +++ b/agent/lagent.md @@ -0,0 +1,73 @@ +# Lagnet + +English | [简体中文](lagent_zh-CN.md) + +## What's Lagent? + +Lagent is a lightweight open-source framework that allows users to efficiently build large language model(LLM)-based agents. It also provides some typical tools to augment LLM. The overview of the framework is shown below: + +![image](https://github.com/InternLM/lagent/assets/24351120/cefc4145-2ad8-4f80-b88b-97c05d1b9d3e) + +This document primarily highlights the basic usage of Lagent. For a comprehensive understanding of the toolkit, please refer to [examples](https://github.com/InternLM/lagent/tree/main/examples) for more details. + +## Installation + +Install with pip (Recommended). + +```bash +pip install lagent +``` + +Optionally, you could also build Lagent from source in case you want to modify the code: + +```bash +git clone https://github.com/InternLM/lagent.git +cd lagent +pip install -e . +``` + +## Run ReAct Web Demo + +```bash +# You need to install streamlit first +# pip install streamlit +streamlit run examples/react_web_demo.py +``` + +Then you can chat through the UI shown as below + +![image](https://github.com/InternLM/lagent/assets/24622904/3aebb8b4-07d1-42a2-9da3-46080c556f68) + +## Run a ReAct agent with InternLM2-Chat + +**NOTE:** If you want to run a HuggingFace model, please run `pip install -e .[all]` first. + +```python +# Import necessary modules and classes from the "lagent" library. +from lagent.agents import ReAct +from lagent.actions import ActionExecutor, GoogleSearch, PythonInterpreter +from lagent.llms import HFTransformer + +# Initialize the HFTransformer-based Language Model (llm) and provide the model name. +llm = HFTransformer('internlm/internlm2-chat-7b') + +# Initialize the Google Search tool and provide your API key. +search_tool = GoogleSearch(api_key='Your SERPER_API_KEY') + +# Initialize the Python Interpreter tool. +python_interpreter = PythonInterpreter() + +# Create a chatbot by configuring the ReAct agent. +chatbot = ReAct( + llm=llm, # Provide the Language Model instance. + action_executor=ActionExecutor( + actions=[search_tool, python_interpreter] # Specify the actions the chatbot can perform. + ), +) +# Ask the chatbot a mathematical question in LaTeX format. +response = chatbot.chat('若$z=-1+\sqrt{3}i$,则$\frac{z}{{z\overline{z}-1}}=\left(\ \ \right)$') + +# Print the chatbot's response. +print(response.response) # Output the response generated by the chatbot. +>>> $-\\frac{1}{3}+\\frac{{\\sqrt{3}}}{3}i$ +``` diff --git a/agent/lagent_zh-CN.md b/agent/lagent_zh-CN.md new file mode 100644 index 0000000..0365969 --- /dev/null +++ b/agent/lagent_zh-CN.md @@ -0,0 +1,73 @@ +# Lagnet + +[English](lagent.md) | 简体中文 + +## 简介 + +[Lagent](https://github.com/InternLM/lagent) 是一个轻量级、开源的基于大语言模型的智能体（agent）框架，支持用户快速地将一个大语言模型转变为多种类型的智能体，并提供了一些典型工具为大语言模型赋能。它的整个框架图如下: + +![image](https://github.com/InternLM/lagent/assets/24351120/cefc4145-2ad8-4f80-b88b-97c05d1b9d3e) + +本文主要介绍 Lagent 的基本用法。更全面的介绍请参考 Lagent 中提供的 [例子](https://github.com/InternLM/lagent/tree/main/examples)。 + +## 安装 + +通过 pip 进行安装 (推荐)。 + +```bash +pip install lagent +``` + +同时，如果你想修改这部分的代码，也可以通过以下命令从源码编译 Lagent: + +```bash +git clone https://github.com/InternLM/lagent.git +cd lagent +pip install -e . +``` + +## 运行一个 ReAct 智能体的网页样例 + +```bash +# 需要确保已经安装 streamlit 包 +# pip install streamlit +streamlit run examples/react_web_demo.py +``` + +然后你就可以在网页端和智能体进行对话了，效果如下图所示 + +![image](https://github.com/InternLM/lagent/assets/24622904/3aebb8b4-07d1-42a2-9da3-46080c556f68) + +## 用 InternLM-Chat 构建一个 ReAct 智能体 + +**注意：**如果你想要启动一个 HuggingFace 的模型，请先运行 pip install -e .[all]。 + +```python +# Import necessary modules and classes from the "lagent" library. +from lagent.agents import ReAct +from lagent.actions import ActionExecutor, GoogleSearch, PythonInterpreter +from lagent.llms import HFTransformer + +# Initialize the HFTransformer-based Language Model (llm) and provide the model name. +llm = HFTransformer('internlm/internlm-chat-7b-v1_1') + +# Initialize the Google Search tool and provide your API key. +search_tool = GoogleSearch(api_key='Your SERPER_API_KEY') + +# Initialize the Python Interpreter tool. +python_interpreter = PythonInterpreter() + +# Create a chatbot by configuring the ReAct agent. +chatbot = ReAct( + llm=llm, # Provide the Language Model instance. + action_executor=ActionExecutor( + actions=[search_tool, python_interpreter] # Specify the actions the chatbot can perform. + ), +) +# Ask the chatbot a mathematical question in LaTeX format. +response = chatbot.chat('若$z=-1+\sqrt{3}i$,则$\frac{z}{{z\overline{z}-1}}=\left(\ \ \right)$') + +# Print the chatbot's response. +print(response.response) # Output the response generated by the chatbot. +>>> $-\\frac{1}{3}+\\frac{{\\sqrt{3}}}{3}i$ +``` diff --git a/agent/pal_inference.md b/agent/pal_inference.md new file mode 100644 index 0000000..851f0d8 --- /dev/null +++ b/agent/pal_inference.md @@ -0,0 +1,62 @@ +# Inference on GSM8K with PAL in InternLM-Chat + +English | [简体中文](pal_inference_zh-CN.md) + +Utilize [PAL](https://github.com/reasoning-machines/pal) paradigm inference on the [GSM8K](https://huggingface.co/datasets/gsm8k) dataset, enabling the model to write code and execute it through the Python interpreter to solve mathematical problems. The usage is as follows: + +```bash +python pal_inference.py \ + \ + \ + [--dataset ] \ + [--max_length ] \ + [--top_p ] \ + [--eoh ] \ + [--eoa ] \ + [--eos ] \ + [--temperature ] \ + [--time_out