openai4rs

简体中文 | English

简介

openai4rs是一个非官方实现的基于tokio和reqwest的异步与大模型供应商以openai规范交互的rust实现的crate。

使用指南

1. chat聊天

• 非流式

use openai4rs::{OpenAI, chat_request, user};

#[tokio::main]
async fn main() {
    let base_url = "your base_url";
    let api_key = "your api_key";
    let client = OpenAI::new(api_key, base_url);
    let messages = vec![user!("hello")];
    let chat_completion = client
        .chat()
        .create(chat_request("your model name", &messages))
        .await
        .unwrap();
    println!("{:#?}", chat_completion);
}

• 流式

use futures::StreamExt;
use openai4rs::{OpenAI, chat_request, user};

#[tokio::main]
async fn main() {
    let base_url = "your base_url";
    let api_key = "your api_key";
    let client = OpenAI::new(api_key, base_url);
    let messages = vec![user!("hello")];
    let mut stream = client
        .chat()
        .create_stream(chat_request("your model name", &messages))
        .await
        .unwrap();
    while let Some(result) = stream.next().await {
        let chat_completion_chunk = result.unwrap();
        println!("{:#?}", chat_completion_chunk);
    }
}

• 使用工具

use futures::StreamExt;
use openai4rs::{ChatCompletionToolParam, OpenAI, chat_request, user};

#[tokio::main]
async fn main() {
    let base_url = "your base_url";
    let api_key = "your api_key";
    let client = OpenAI::new(api_key, base_url);
    // 定义工具
    let tools = vec![ChatCompletionToolParam::function(
        "get_time",
        "获取当前时间",
        serde_json::json!({
            "type": "object",
            "properties": {}
        }),
    )];

    let messages = vec![user!("现在几点了?")];
    let mut stream = client
        .chat()
        .create_stream(
            chat_request("your model name", &messages)
                .tools(tools.clone())
                .tool_choice(openai4rs::ToolChoice::Auto), // 选择工具模式
        )
        .await
        .unwrap();
    while let Some(result) = stream.next().await {
        match result {
            Ok(chunk) => {
                println!("{:#?}", chunk);
            }
            Err(err) => {
                println!("{:#?}", err);
            }
        }
    }
}

• 思考模型

模型必须支持思考功能, 例如qwen的qwq-32b

use futures::StreamExt;
use openai4rs::{OpenAI, chat_request, user};

#[tokio::main]
async fn main() {
    let base_url = "your base_url";
    let api_key = "your api_key";
    let client = OpenAI::new(api_key, base_url);
    let messages = vec![user!("hello")];
    let mut stream = client
        .chat()
        .create_stream(chat_request("your model name", &messages))
        .await
        .unwrap();
    while let Some(result) = stream.next().await {
        let chat_completion_chunk = result.unwrap();
        for choice in chat_completion_chunk.choices.iter() {
            if choice.delta.is_reasoning() {
                println!("## REASONING\n{}", choice.delta.get_reasoning_str())
            }
            if let Some(content) = &choice.delta.content {
                if !content.is_empty() {
                    println!("## CONTENT\n{}", content);
                }
            }
        }
    }
}