Anthropic Tools: Say Goodbye to Traditional Function Calling

Summary

这个视频教程向观众展示了如何使用Anthropic工具简化与Anthropic AI模型Claude进行功能调用的过程。首先,演示了如何从GitHub克隆Anthropic工具库并安装必要的依赖项。然后,通过设置工具名称、描述和参数,创建了一个简单的加法工具。最后,演示了如何将此工具分配给Claude,并要求其解决一个涉及水果计数的问题。总的来说,该教程向观众介绍了使用Anthropic工具的基本流程,这种方式比传统的功能调用更加简单高效。

Takeaways

• 🤖 本视频介绍了 Anthropic 工具的使用, 可以简化函数调用的过程。
• 📝 首先需要克隆 github.com/anthropic/anthropic-tools 仓库, 安装相关依赖。
• ✏️ 接着在 app.py 中从 tools 包导入 base_tool 和 tool_user。
• ➕ 定义了一个加法工具, 可以将两个数相加。
• 🔖 为工具添加名称、描述和参数说明, 以帮助 AI 理解工具的功能。
• 🧩 将定义好的工具添加到 tool_user 中。
• ✉️ 在 messages 变量中提出一个需要计算的问题。
• 🚀 调用 tool_user.use_tools 来执行任务, 传入 messages 和执行模式。
• 💻 运行 app.py 后可以看到正确的计算结果。
• 🌐 其他示例工具包括天气、转换器、维基百科搜索等, 可供进一步尝试。

Q & A

什么是Anthropic工具?

Anthropic工具是一种简化函数调用过程的方法,可以让你直接将自己的函数或知识库集成到大型语言模型中,而不需要处理繁琐的XML标记。

为什么需要Anthropic工具?

大型语言模型在处理一些任务时表现不佳,比如数学计算。通过Anthropic工具,我们可以将自定义功能集成到语言模型中,弥补其不足。

使用Anthropic工具有哪些步骤?

一共有三个步骤:1.创建工具及其描述 2.将工具添加到Anthropic API 3.要求语言模型执行任务

视频中演示的是什么样的工具?

视频中演示的是一个加法计算工具,可以将两个数字相加。

Anthropic工具是如何与语言模型交互的?

通过ToolUser包中的代码,Anthropic工具会将请求发送到API,语言模型会根据工具的功能给出响应。

Anthropic工具有哪些优势?

Anthropic工具使得集成自定义功能变得非常简单,无需处理复杂的XML标记,可以更高效地利用语言模型的能力。

除了加法工具,视频中还提到了哪些其他工具?

视频中还提到了天气工具、信息API转换工具、计算器工具、维基百科搜索工具、Brave搜索工具、ElasticSearch工具和SQL工具。

如何创建自定义的Anthropic工具?

你需要从basetools包导入BaseTools类,定义工具的名称、描述和参数,然后将其添加到ToolUser中。

为什么大型语言模型在处理某些任务时表现不佳?

大型语言模型通常在某些特定领域或需要外部数据源的任务中表现不佳,这就是需要Anthropic工具来弥补不足的原因。

除了提供的示例,Anthropic工具还可以用于哪些其他场景?

Anthropic工具可以用于各种需要调用外部API、数据库或其他服务的场景,为语言模型提供更强大的功能扩展。

Outlines

00:00

🔑 介紹 Anthropic 工具

本段詳細介紹了 Anthropic 工具的概念,並說明了它如何簡化函數調用的過程。作者還提到,他將逐步講解如何創建 Anthropic 工具、設置工具描述、將工具添加到 Anthropic API 中以及執行任務。同時,作者邀請觀眾訂閱他的 YouTube 頻道以獲取更多關於人工智能的視頻內容。最後,他演示了如何設置 Python 開發環境、導入所需的包以及創建一個新文件 app.py 供後續編碼使用。

05:01

🏁 總結和呼籲參與

本段總結了前面介紹的 Anthropic 工具,並讚揚了它簡化函數調用的優勢。作者提到 Anthropic 還提供了其他工具樣例供用戶嘗試,如天氣工具、消息 API 轉換工具、計算器工具等。最後,作者鼓勵觀眾在評論區分享對本教程的反饋,並宣佈了他計劃製作更多類似的視頻內容,敦促觀眾訂閱他的頻道,對視頻點贊。

Keywords

💡anthropic 工具

该视频主要介绍了 Anthropic 公司新推出的 AI 工具。Anthropic 工具是一种简化人工智能模型功能调用的新方式,它能够通过定义工具并添加到 Anthropic API 中,从而直接与语言模型交互以完成特定任务。例如,视频中展示了如何创建一个"加法工具"用于计算两个数字的和。这种工具使用更加简单直观,而无需手动处理 XML 标签和响应解析等复杂步骤。

💡功能调用

功能调用是指调用外部函数或工具来完成特定任务。在人工智能领域,功能调用通常指将语言模型与外部知识库、计算工具或服务集成,以扩展模型的能力。视频中提到,以前的功能调用方式需要手动处理 XML 标签和解析响应,而新的 Anthropic 工具则简化了这一过程。

💡cloe

Cloe 是 Anthropic 公司开发的人工智能语言模型。正如视频所述,之前的教程演示了如何在 Cloe 中进行功能调用。而新的 Anthropic 工具则提供了一种更简单的方式来与 Cloe 模型集成外部工具和功能。

💡api

API(应用程序编程接口)是软件系统与外部工具或服务进行交互的接口。视频演示了如何通过导入并设置 Anthropic API 密钥,然后将定义好的工具添加到 API 中,从而实现与语言模型 Cloe 的无缝集成。这种基于 API 的集成方式提供了灵活性和可扩展性。

💡git克隆

Git 克隆是从远程代码库中复制代码到本地的过程。视频中的第一步就是运行 "git clone" 命令从 Anthropic 的 GitHub 仓库中获取 Anthropic 工具的源代码。这是开始使用该工具的必要步骤。

💡python 环境

Python 环境是指运行 Python 代码所需的依赖库和配置。视频演示了如何使用 conda 创建一个新的 Python 3.11 环境,并将其命名为"anthropic"。这确保了代码在干净隔离的环境中运行,避免与系统上其他 Python 包产生冲突。

💡安装依赖

安装依赖指的是安装运行代码所需的外部 Python 包。在视频中, "pip install -r requirements.txt" 命令用于一次性安装 Anthropic 工具所依赖的所有第三方 Python 包。这确保了代码可以正常运行。

💡工具描述

工具描述是对将要添加到 Anthropic API 中的工具进行的文字描述。在视频代码中,工具描述"Add one number to another number returning a + b"解释了加法工具的功能,使语言模型能够理解该工具的用途。合理的工具描述对于模型正确使用工具至关重要。

💡自动执行模式

自动执行模式是 Anthropic 工具中的一种运行方式。在视频示例中,代码通过"execution_mode='automatic'"参数指定了自动执行模式。这意味着 Cloe 模型将自动识别出需要使用加法工具来计算的情况,而无需人为指示。自动模式提高了使用效率。

💡github

GitHub 是一个面向开源及私有软件项目的托管平台,被广泛用于代码存储、版本控制和协作开发。视频中提到从 Anthropic 的 GitHub 仓库 "github.com/anthropics/anthropic-tools" 克隆工具代码,这说明工具源代码是开源和公开的。

Highlights

This video demonstrates how to use Anthropic Tools to simplify the process of function calling with Claude.

The first step is to create a tool by defining its functionality, such as an addition tool that adds two numbers.

The second step is to create a description for the tool to help Claude understand what the tool does.

The third step is to assign the tool to Claude and provide a task that requires the tool's functionality.

The video shows how to set up the project environment, including cloning the Anthropic Tools repository, creating a virtual environment, and installing the required packages.

The video demonstrates running the code with the addition tool and how Claude uses the tool to solve a word problem involving addition.

The video highlights that Anthropic Tools simplify the process of function calling by handling the heavy lifting of parsing XML values, making API calls, and defining the model.

The video mentions that Anthropic Tools provide various example tools, such as weather, message API converter, calculator, Wikipedia search, Brave search, Elasticsearch, and SQL tools.

The video encourages viewers to try the example tools and provide feedback in the comments.

The video is part of a series on Artificial Intelligence, and the creator regularly creates videos on this topic.

The video emphasizes the importance of subscribing, clicking the bell icon, and liking the video to stay updated and support the creator's efforts.

The video provides step-by-step instructions for setting up the project environment and running the code with the addition tool.

The video explains how the code is structured, with the base_tool package handling the tool functionality and the tool_user package interacting with the Claude API.

The video demonstrates how to define the tool parameters, including the expected data types.

The video showcases the log output that displays each time Claude calls the addition tool, providing transparency into the tool's usage.

Transcripts

this is amazing now we have anthropic

tools from anthropic so what is

anthropic tool this is similar to

function calling but simplifies the

process of function calling I've already

covered how to do function calling in

anthropic Cloe in my previous video

which I will link that in the

description below in function calling

you can see a lot of XML tags we need to

think about we need to pause these XML

tags then calculate to get a final

response but this anthropic tool will

simplify all that work that's exactly

what we're going to see today let's get

[Music]

started hi everyone I'm really excited

to show you about anthropic Tools in

this we are going to cover how to create

anthropic tool set tool description add

the tool to anthropic API and finally

perform a task I'm going to take you

through step by step but before that I

regularly create videos in regards to

Artificial Intelligence on my YouTube

channel so do subscribe and click the

Bell icon to stay stay tuned make sure

you click the like button so this video

can be helpful for many others like you

first git clone github.com

anthropics anthropic tools and then

click enter next navigate to the

anthropic tools folder now cond create

hyphen and anthropic python 311 and then

click enter Now cond activate anthropic

and click enter export your anthropic AP

key like this and then click enter now

pip install hyphen requirements.txt and

then click enter now let's create a file

called app.py and let's open it so if

you see anthropic tools folder this is

how it will look like in the tool use

package you have tool user and inside

the tools folder we have base tool so

these are the two packages which we are

going to use base tool and Tool user

tool user calls the API using base tool

we are going to create a tool so inside

app.py from Tool use package tool base

tool and import base tool next importing

tool user so the first step defining the

tool we are going to create a addition

tool this is going to add two numbers

together which you can see here return a

plus b a simple tool so as you probably

know function calling is integrating

your own function or own tool own

knowledge base own rag with a lar

language model now second step we need

to create description for the tool so

this is used by the L language model to

understand what this tool does so name

of the tool is perform addition tool

description is add one number to another

number returning a plus b next tool

parameters A and B both float which

means number in simple terms now we

defining the addition tool and pausing

the addition tool name addition tool

description and addition tool parameters

all three together like this now the

third step is to assign the tool and ask

clo to perform a task so as mentioned

before tool user is is where the API

request to Cloe happens so we defining

that and adding the tool here next we

are giving a task it is listed in the

messages variable here you can see as a

user that is me I'm asking Cloe John has

three apples Maggie has nine apples Tim

has four bananas if John gives Maggie

one apple and Tim gives John two bananas

how much total fruits does each person

have so as you properly know large

language model are not good at

mathematical calculation so we need to

have a tool to perform this task so

that's why we added this addition tool

now finally we are going to initiate by

calling tool user.use tools and passing

the messages execution mode is automatic

that's it only three section first we

created a tool this could be your own

tool our own software then we Define a

description for the tool thirdly we

assign that tool and asking Cloe to

complete this task now we are going to

run this code in your terminal Python

app.py and then click enter now you can

see it's adding three and minus one so

I've added the log here so that's what

it's printing here so each and every

single time Cloe is calling this tool

that is the time it is getting printed

here so now it's calculating using this

calculator tool and finally here is the

answer in summary John has four total

fruits two apples and two bananas Maggie

has 10 apples Tim has two bananas that

is correct using anthropic tools is the

most simple approach in regards to do

function calling in anthropic clo this

is just a basic example now to show the

files you can see in this tool use

package this tool user is doing all the

heavy lifting of parsing those XML

values which we saw previous ly making

the API call defining the model and much

more you also have many other example

tools which you can try which is the

weather tool messages API converter tool

calculator tool Wikipedia search tool

Brave Search tool elastic search tool

SQL tool now you can try and let me know

in the comments below what you think

about this I'm really excited about this

I'm going to create more videos similar

to this so stay tuned I hope you like

this video do like sh subscribe and

thanks for watching