kajin-leo / demo-baiduqianfan-chatbot

A demo for ChatBot based on BaiduQianfan LLM and LangChain.

中文|English

中文Readme

这个Demo是基于LangChain和百度千帆模型搭建的聊天机器人，其中兼顾了Completion模式与RAG模式。Demo来源于作者在实习时做的任务，基于公司Wiki库回答用户提问的聊天机器人。

LangChain有利于Demo向不同的大型语言模型进行改造，在参考LangChain官方Tutorial的情况下可以通过较少量的修改来实现。另外，在Demo中使用了Flask库进行包装，以便响应网络请求。

附加文件说明

项目的虚拟环境为.venv目录，在.venv目录下会有一个名为hash.json的文件以及一个data文件夹。hash.json文件是用于实现RAG模式的知识库增量上传的辅助文件。data文件夹下存有知识库数据源的.csv文件。

目前Demo中只使用了CSVLoader来读取知识库，如果有需要的其他格式，可以参考LangChain的Loader介绍进行添加和修改。

开始之前

在main.py的25行至27行，请首先输入您的LangChain API Key和百度千帆AK与SK！

通过网络请求与接口交互

由于作者的实习单位主要使用PHP语言，作者也对B/S开发经验不足，这里只提供PHP语言的参考脚本。

清理数据库与检查数据库内数据量

function callChatBotAPI($operation){ //$operation是clear或check
    $url = "http://localhost:8204/chatbot/{$operation}";
    $options = array(
        'http' => array(
            'header' => 'Content-type: application/json\r\n',
            'method' => 'POST'
        )
    );
    $context = stream_context_create($options);
    $result = file_get_contents($url, false, $context);
    if ($result === FALSE) {
        throw new Exception('Failed to interact with chatbot');
    }
    if ($operation == 'check') {
        return json_decode($result, true)["library_name"] . json_decode($result, true)["library_count"];
    }
    return json_decode($result, true)['output'];
}

上传知识库

function callChatBotAPI(){
    $url = 'http://localhost:8204/chatbot/upload';
    $ch = curl_init($url);
    curl_setopt($ch, CURLOPT_POST, 1);
    curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
    curl_setopt($ch, CURLOPT_WRITEFUNCTION, function ($curl, $data) {
        $output = json_decode($data, true)["output"];
        echo "{$output}\n";
        return strlen($data);
    });
    curl_exec($ch);
    curl_close($ch);
}

Wiki与Completion模式

function callChatBotAPI($operation, $content){ //$operation是wiki或completion
    $ch = curl_init($url);
    $context = json_encode(array('content' => $content));
    curl_setopt($ch, CURLOPT_POST, 1);
    curl_setopt($ch, CURLOPT_POSTFIELDS, $context);
    curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
    curl_setopt($ch, CURLOPT_HTTPHEADER, [
        'Content-Type: application/json',
        'Accept: text/event-stream'
    ]);
    curl_setopt($ch, CURLOPT_HEADER, true);
    curl_setopt($ch, CURLOPT_WRITEFUNCTION, function ($curl, $data) {
        $output = json_decode($data, true);
        if($output != null && $output["status"] == "active") {
            echo "{$output["output"]}";
        }
        return strlen($data);
    });
    curl_exec($ch);
    curl_close($ch);
}

English Readme

This demo is a ChatBot based on LangChain and BaiduQianfan LLM, which contains both Completion mode and RAG mode. The Demo was originated from a task during the internship of the author, which was to construct a ChatBot based on the Wiki Library of the company.

LangChain makes it much easier to modify for different LLM models. With the help of LangChain Official Tutorial, it's possible to change the demo to other LLMs through only a little bit of codings. Additionally, 'Flask' was used to pack the demo for responding to HTTP Requests.

Descriptions for Additional Files

There is a directory called .venv to work as the virtual environment, containing a file named as hash.json and a directory named as data. hash.json is used to assist with uploading the wiki library incrementally. data directory stores wiki library files in .csv filename extension. So far, the demo only imports and uses CSVLoader to load the wiki library. If any other file format is needed, you can take LangChain Loaders Introduction as reference to modify the demo.

Before You Start

Please specify your LangChain API Key and Baidu Qianfan AK & SK in main.py from Line 25 to Line 27.

Using HTTP Requests

Since the author is not familiar with B/S developing and PHP was mainly used in the company where the author had internship, all the following referral scripts are wrote in PHP.

Clear the Vector Database & Check the Vector Database

function callChatBotAPI($operation){ //$operation should only be in 'clear' or 'check'
    $url = "http://localhost:8204/chatbot/{$operation}";
    $options = array(
        'http' => array(
            'header' => 'Content-type: application/json\r\n',
            'method' => 'POST'
        )
    );
    $context = stream_context_create($options);
    $result = file_get_contents($url, false, $context);
    if ($result === FALSE) {
        throw new Exception('Failed to interact with chatbot');
    }
    if ($operation == 'check') {
        return json_decode($result, true)["library_name"] . json_decode($result, true)["library_count"];
    }
    return json_decode($result, true)['output'];
}

Upload Wiki Library

function callChatBotAPI(){
    $url = 'http://localhost:8204/chatbot/upload';
    $ch = curl_init($url);
    curl_setopt($ch, CURLOPT_POST, 1);
    curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
    curl_setopt($ch, CURLOPT_WRITEFUNCTION, function ($curl, $data) {
        $output = json_decode($data, true)["output"];
        echo "{$output}\n";
        return strlen($data);
    });
    curl_exec($ch);
    curl_close($ch);
}

Chat using Wiki & Completion Modes

function callChatBotAPI($operation, $content){ //$operation should only be in 'wiki' or 'completion'
    $ch = curl_init($url);
    $context = json_encode(array('content' => $content));
    curl_setopt($ch, CURLOPT_POST, 1);
    curl_setopt($ch, CURLOPT_POSTFIELDS, $context);
    curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
    curl_setopt($ch, CURLOPT_HTTPHEADER, [
        'Content-Type: application/json',
        'Accept: text/event-stream'
    ]);
    curl_setopt($ch, CURLOPT_HEADER, true);
    curl_setopt($ch, CURLOPT_WRITEFUNCTION, function ($curl, $data) {
        $output = json_decode($data, true);
        if($output != null && $output["status"] == "active") {
            echo "{$output["output"]}";
        }
        return strlen($data);
    });
    curl_exec($ch);
    curl_close($ch);
}