284 lines
8.5 KiB
Markdown
284 lines
8.5 KiB
Markdown
# LLMs 大语言模型
|
||
|
||
Agents-Flex 提供了关于大语言模型的抽象实现接口 `Llm.java`，它支持 `chat` 以及 `chatStream` 两种不同的对话方式。
|
||
|
||
针对不同的厂商，Agents-Flex 提供了不同的实现类以及通信协议，其中通信协议包括了 `HTTP`、`SSE` 以及 `WebSocket` 等客户端。
|
||
|
||
## 大模型支持
|
||
|
||
目前，Agents-Flex 已支持以下的大语言模型：
|
||
|
||
- OpenAI（ChatGPT，以及所有的兼容 OpenAI 接口的大模型）
|
||
- ChatGLM（智普大模型）
|
||
- Coze （调用 Coze 的智能体）
|
||
- DeepSeek
|
||
- Gitee AI
|
||
- Moonshot（月之暗面）
|
||
- Ollama（通过 Ollama 部署的所有大模型）
|
||
- Qianfan（百度千帆部署的大模型）
|
||
- Qwen（千问大模型，阿里云百炼平台部署的大模型）
|
||
- Spark（星火大模型）
|
||
|
||
|
||
## 简单对话
|
||
```java
|
||
public static void main(String[] args) {
|
||
OpenAILlmConfig config = new OpenAILlmConfig();
|
||
|
||
// 设置你的 OpenAI API Key
|
||
config.setApiKey("sk-rts5NF6n*******");
|
||
|
||
Llm llm = new OpenAILlm(config);
|
||
String response = llm.chat("请问你叫什么名字");
|
||
|
||
System.out.println(response);
|
||
}
|
||
```
|
||
|
||
## 流式对话
|
||
|
||
```java
|
||
public static void main(String[] args) {
|
||
|
||
OpenAILlmConfig config = new OpenAILlmConfig();
|
||
// 设置你的 OpenAI API Key
|
||
config.setApiKey("sk-rts5NF6n*******");
|
||
|
||
Llm llm = new OpenAILlm(config);
|
||
|
||
llm.chatStream("你叫什么名字", new StreamResponseListener() {
|
||
|
||
@Override
|
||
public void onMessage(ChatContext context, AiMessageResponse response) {
|
||
System.out.println(response.getMessage().getContent());
|
||
}
|
||
|
||
});
|
||
}
|
||
```
|
||
|
||
## 流式对话之停止对话
|
||
|
||
```java
|
||
public static void main(String[] args) {
|
||
OpenAILlmConfig config = new OpenAILlmConfig();
|
||
|
||
// 设置你的 OpenAI API Key
|
||
config.setApiKey("sk-rts5NF6n*******");
|
||
|
||
Llm llm = new OpenAILlm(config);
|
||
llm.chatStream("你叫什么名字", new StreamResponseListener() {
|
||
@Override
|
||
public void onMessage(ChatContext context, AiMessageResponse response) {
|
||
System.out.println(response.getMessage().getContent());
|
||
|
||
//停止对话
|
||
context.getClient().stop();
|
||
}
|
||
});
|
||
}
|
||
```
|
||
|
||
|
||
## 流式对话之更多的监听
|
||
|
||
```java
|
||
public static void main(String[] args) {
|
||
OpenAILlmConfig config = new OpenAILlmConfig();
|
||
|
||
// 设置你的 OpenAI API Key
|
||
config.setApiKey("sk-rts5NF6n*******");
|
||
|
||
Llm llm = new OpenAILlm(config);
|
||
llm.chatStream("你叫什么名字", new StreamResponseListener() {
|
||
@Override
|
||
public void onMessage(ChatContext context, AiMessageResponse response) {
|
||
AiMessage message = response.getMessage();
|
||
System.out.println(message.getContent());
|
||
}
|
||
|
||
@Override
|
||
public void onStart(ChatContext context) {
|
||
System.out.println("开始");
|
||
}
|
||
|
||
@Override
|
||
public void onStop(ChatContext context) {
|
||
System.out.println("结束");
|
||
}
|
||
|
||
@Override
|
||
public void onFailure(ChatContext context, Throwable throwable) {
|
||
System.out.println("错误");
|
||
}
|
||
});
|
||
}
|
||
```
|
||
|
||
## 图片识别对话
|
||
|
||
```java
|
||
public static void main(String[] args) {
|
||
OpenAILlmConfig config = new OpenAILlmConfig();
|
||
|
||
// 设置你的 OpenAI API Key
|
||
config.setApiKey("sk-5gqOcl*****");
|
||
config.setModel("gpt-4-turbo");
|
||
|
||
|
||
Llm llm = new OpenAILlm(config);
|
||
ImagePrompt prompt = new ImagePrompt("What's in this image?");
|
||
prompt.setImageUrl("https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg");
|
||
|
||
//或者 prompt.setImageFile(new File("/your-image-path.png"))
|
||
//或者 prompt.setImageStream(imageInputStream)
|
||
//或者 prompt.setImageBase64("image base64 data....")
|
||
|
||
AiMessageResponse response = llm.chat(prompt);
|
||
System.out.println(response);
|
||
}
|
||
```
|
||
|
||
## 方法调用（Function Calling）
|
||
|
||
```java
|
||
public static void main(String[] args) {
|
||
OpenAILlmConfig config = new OpenAILlmConfig();
|
||
config.setApiKey("sk-rts5NF6n*******");
|
||
|
||
OpenAILlm llm = new OpenAILlm(config);
|
||
|
||
FunctionPrompt prompt = new FunctionPrompt("今天北京的天气怎么样", WeatherFunctions.class);
|
||
AiMessageResponse response = llm.chat(prompt);
|
||
|
||
System.out.println(response.callFunctions());
|
||
// "Today it will be dull and overcast in 北京"
|
||
}
|
||
```
|
||
|
||
`WeatherFunctions.class` 代码如下：
|
||
|
||
```java
|
||
public class WeatherFunctions {
|
||
|
||
@FunctionDef(name = "get_the_weather_info", description = "get the weather info")
|
||
public static String getWeatherInfo( @FunctionParam(name = "city", description = "the city name") String name) {
|
||
return "Today it will be dull and overcast in " + name;
|
||
}
|
||
}
|
||
```
|
||
|
||
|
||
## 历史对话
|
||
|
||
```java
|
||
public static void main(String[] args) {
|
||
OpenAILlmConfig config = new OpenAILlmConfig();
|
||
config.setApiKey("sk-rts5NF6n*******");
|
||
|
||
OpenAILlm llm = new OpenAILlm(config);
|
||
|
||
//第一步：创建一个 HistoriesPrompt
|
||
HistoriesPrompt prompt = new HistoriesPrompt();
|
||
|
||
System.out.println("您想问什么？");
|
||
Scanner scanner = new Scanner(System.in);
|
||
String userInput = scanner.nextLine();
|
||
|
||
while (userInput != null) {
|
||
|
||
// 第二步：将用户输入添加到 HistoriesPrompt 中
|
||
prompt.addMessage(new HumanMessage(userInput));
|
||
|
||
// 第三步：调用 chatStream 方法，进行对话
|
||
llm.chatStream(prompt, (context, response) -> {
|
||
System.out.println(">>>> " + response.getMessage().getContent());
|
||
});
|
||
|
||
userInput = scanner.nextLine();
|
||
}
|
||
}
|
||
```
|
||
|
||
关于 HistoriesPrompt 更多的配置：
|
||
|
||
```java
|
||
HistoriesPrompt prompt = new HistoriesPrompt();
|
||
|
||
//设置系统消息
|
||
prompt.setSystemMessage(new SystemMessage('你是一个数据库开发工程师....'));
|
||
|
||
//设置最大历史消息数量
|
||
prompt.setMaxAttachedMessageCount(10);
|
||
|
||
//设置是否开启历史消息截断
|
||
prompt.setHistoryMessageTruncateEnable(true);
|
||
|
||
//设置历史消息截断长度
|
||
prompt.setHistoryMessageTruncateLength(1000);
|
||
|
||
//自定义历史消息截断处理器
|
||
prompt.setHistoryMessageTruncateProcessor(...);
|
||
|
||
//设置历史消息存储器
|
||
prompt.setMemory(...);
|
||
```
|
||
|
||
## 历史对话 + 方法调用
|
||
|
||
```java
|
||
public static void main(String[] args) {
|
||
OpenAILlmConfig config = new OpenAILlmConfig();
|
||
config.setApiKey("sk-rts5NF6n*******");
|
||
|
||
OpenAILlm llm = new OpenAILlm(config);
|
||
|
||
//第一步：创建一个 HistoriesPrompt
|
||
HistoriesPrompt prompt = new HistoriesPrompt();
|
||
|
||
System.out.println("您想问什么？");
|
||
Scanner scanner = new Scanner(System.in);
|
||
String userInput = scanner.nextLine();
|
||
|
||
while (userInput != null) {
|
||
|
||
// 第二步：创建 HumanMessage，并添加方法调用
|
||
HumanMessage humanMessage = new HumanMessage(userInput);
|
||
humanMessage.addFunctions(WeatherFunctions.class);
|
||
|
||
// 第三步：将 HumanMessage 添加到 HistoriesPrompt 中
|
||
prompt.addMessage(humanMessage);
|
||
|
||
// 第四步：调用 chatStream 方法，进行对话
|
||
llm.chatStream(prompt, new StreamResponseListener() {
|
||
@Override
|
||
public void onMessage(ChatContext context, AiMessageResponse response) {
|
||
boolean functionCall = response.isFunctionCall();
|
||
if (functionCall) {
|
||
System.out.println("do func >>> ");
|
||
StringBuilder text = new StringBuilder("调用工具结果如下：\n");
|
||
List<FunctionCaller> callers = response.getFunctionCallers();
|
||
for (FunctionCaller caller : callers) {
|
||
String name = caller.getFunction().getName();
|
||
Object callRes = caller.call();
|
||
text.append("调用[").append(name).append("]的结果为：").append(callRes).append("\n");
|
||
}
|
||
HumanMessage msg = new HumanMessage(text.toString());
|
||
prompt.addMessage(msg);
|
||
llm.chatStream(prompt, new StreamResponseListener() {
|
||
@Override
|
||
public void onMessage(ChatContext context, AiMessageResponse response) {
|
||
System.out.println("after func >>>> " + response.getMessage().getContent());
|
||
}
|
||
});
|
||
} else {
|
||
System.out.println("normal >>>> " + response.getMessage().getContent());
|
||
}
|
||
}
|
||
});
|
||
|
||
userInput = scanner.nextLine();
|
||
}
|
||
}
|
||
```
|