Programmatically add website content to OpenAI with C#

题意:使用 C# 以编程方式将网站内容添加到 OpenAI。

问题背景:

Our goal is to have a ChatGPT answer questions about our website's content.

我们的目标是让 ChatGPT 回答关于我们网站内容的问题。

We are trying to integrate something similar to a tutorial by Open AI:

我们正在尝试集成类似于 OpenAI 提供的教程的东西

How to build an AI that can answer questions about your website. I've been converting the code from Python to C# and all has been pretty successful so far. In detail we are storing the files in Azure Storage blobs.

如何构建一个可以回答关于你的网站问题的 AI。我已经将代码从 Python 转换为 C#,到目前为止一切都非常成功。详细来说,我们将文件存储在 Azure Storage blobs 中。

As I've gotten to the end of this, I noticed that for every question the embeddings are being sent as the context (pardon if that's not correctly worded). Our website has 5000+ pages with lots of content so I feel like this is not correct to what we want.

在完成这个过程后,我注意到每个问题的上下文都被发送了嵌入(如果表述不准确请见谅)。我们的网站有超过 5000 页的内容,所以我觉得这样做可能不符合我们的需求。

Is there a way in C# to give (or maybe it's called 'train') our model this data once so that questions do not need to continuously provide it for every question? We are using OpenAI and not Azure OpenAI but all the samples I find concerning using my own data are using Azure OpenAI and Azure Cognitive Search, and even then it seems like they are feeding the data for each question.

在 C# 中是否有办法一次性给我们的模型提供这些数据(或许称之为"训练"),这样提问时就不需要每次都提供这些数据?我们使用的是 OpenAI 而不是 Azure OpenAI,但我找到的所有关于使用自己数据的示例都是关于 Azure OpenAI 和 Azure Cognitive Search 的,即便如此,它们似乎还是在每个问题中都提供数据

cs 复制代码
import Azure.AI.OpenAI

var apiKey = "<my api key>";
var client = new OpenAIClient(apiKey);
var prompt = "<question about website>";

var options = new ChatCompletionOptions {
    Messages = {
        new ChatMessage(ChatRole.User, prompt),
    },
    /*
    Not currently working for some reason, further inspection needed;
    - Is it because we need Azure OpenAI instead?;
    Receiving the following error:
        
        - Azure.RequestFailedException: Invalid URL (POST /v1/extensions/chat/completions)
        - Status: 404 (Not Found)


    AzureExtensionsOptions = new AzureChatExtensionsOptions {
        Extensions = {
            new Azure CognitiveSearchExtensionConfiguration {
                SearchEndpoint = new Uri("<my search url>"),
                IndexName = "<my index name>",
                SearchKey = new AzureKeyCredential("<my search key>"),
            }
        }
    }*/
};

var response = await GetChatCompletionsAsync("gpt-3.5-turbo", options);

foreach (var choice in response.Value.Choices) {
    Console.WriteLine(choice.Message.Content);
}

问题解决:

We finally applied for and started using Azure Open AI and Azure Cognitive Search. The results are huge! Results are now in 2-3 seconds.

我们终于申请并开始使用 Azure OpenAI 和 Azure Cognitive Search。结果非常显著!现在结果只需 2-3 秒

相关推荐
武子康2 小时前
调查研究-203 SpaceX IPO 总览:先别急着讲故事,先把发行事实和信息边界立住
人工智能·openai·agent
怕浪猫3 小时前
第7章 检索增强生成:打造知识库驱动型Agent
aigc·openai·ai编程
doiito4 小时前
【Agent Harness】Gliding Horse 核心设计理念,不跟风开发自己的AI Agent
ai·rust·架构设计·系统设计·ai agent
uccs13 小时前
流式响应的三次进化:EventSource → ReadableStream → TransformStream
openai·ai编程·claude
宅小年17 小时前
Codex Skills 怎么选?我常用的几个推荐给你
openai
机器之心18 小时前
近80年后,埃尔德什经典「拉姆齐数下界」,被三位中国学者首次指数级改进
人工智能·openai
机器之心18 小时前
Nvidia都在点赞的LoopWM世界模型,竟然来自一家中国初创FaceMind?
人工智能·openai
武子康1 天前
调查研究-202 SGLang 深度解析:为什么大模型推理框架不只是“把模型跑起来“
人工智能·openai·agent
怕浪猫1 天前
第6章 多智能体协作:从单兵作战到群体智能
aigc·openai·ai编程
doiito1 天前
【Agent Harness】Gliding Horse 的 L2 作战地图:让多 Agent 协作从“摸黑”变成“透明”
ai·rust·架构设计·系统设计·ai agent