OpenAI Function Call Java Example

From the project: https://github.com/HamaWhiteGG/langchain-java

Please refer to the ChatFunctionTest for the complete test code.

1. Core main process

  • Generates a ChatParameter instance that represents the given class.
  • Invoke the OpenAI API and return the function name and parameters.
  • Execute functions through the FunctionExecutor.

2. Step-by-step introduction

2.1 Declare function parameters and response

java 复制代码
public record Weather(
        @JsonProperty(required = true)
        @JsonPropertyDescription("The city and state, e.g. San Francisco, CA")
        String location,

        @JsonPropertyDescription("The temperature unit")
        WeatherUnit unit
) {}


public enum WeatherUnit {

    CELSIUS,
    FAHRENHEIT
}


@Data
@Builder
public class WeatherResponse {

    public String location;

    public WeatherUnit unit;

    public int temperature;

    public String description;
}


WeatherResponse getCurrentWeather(Weather weather) {
    // mock function
    return WeatherResponse.builder()
            .location(weather.location())
            .unit(weather.unit())
            .temperature(new Random().nextInt(50))
            .description("sunny")
            .build();
}

2.2 Generates a ChatParameter instance that represents the given class

Automatically convert Weather.class to ChatParameter using the ChatParameterUtils.generate method.

java 复制代码
ChatFunction.ChatParameter chatParameter = ChatParameterUtils.generate(Weather.class);

The output of converting chatParameter to JSON String is as follows:

json 复制代码
{
    "type" : "object",
    "properties" : {
        "location" : {
            "type" : "string",
            "description" : "The city and state, e.g. San Francisco, CA"
        },
        "unit" : {
            "type" : "string",
            "description" : "The temperature unit",
            "enum" : [
                "celsius",
                "fahrenheit"
            ]
        }
    },
    "required" : [
        "location"
    ]
}

2.3 Invoke the OpenAI API and return the function name and parameters.

java 复制代码
ChatFunction chatFunction = ChatFunction.builder()
        .name(functionName)
        .description("Get the current weather in a given location")
        .parameters(ChatParameterUtils.generate(Weather.class))
        .build();

Message message = Message.of("What is the weather like in Boston?");

ChatCompletion chatCompletion = ChatCompletion.builder()
        .model("gpt-4")
        .temperature(0)
        .messages(List.of(message))
        .tools(List.of(new Tool(chatFunction)))
        .toolChoice("auto")
        .build();

ChatCompletionResp response = client.createChatCompletion(chatCompletion);
ChatChoice chatChoice = response.getChoices().get(0);

The output of converting chatChoice to JSON String is as follows:

json 复制代码
{
    "index":0,
    "message":{
        "role":"assistant",
        "content":null,
        "tool_calls":[
            {
                "id":"call_aS4LspVVBkE8uIiBSLZWXe6O",
                "type":"function",
                "function":{
                    "name":"get_current_weather",
                    "arguments":"{\n  \"location\": \"Boston, MA\"\n}"
                }
            }
        ]
    },
    "finish_reason":"tool_calls"
}

2.4 Execute functions through the FunctionExecutor.

The code for FunctionExecutor is as follows, allowing the caller to avoid explicit type casting through generics, making the code more concise and elegant.

java 复制代码
public class FunctionExecutor {

    private static final Map<String, Function<?, ?>> FUNCTION_MAP = new HashMap<>(16);

    public static <T> void register(String functionName, Function<T, ?> function) {
        FUNCTION_MAP.put(functionName, function);
    }

    @SuppressWarnings("unchecked")
    public static <T, R> R execute(String functionName, Class<T> clazz, String arguments) throws ClassCastException {
        Function<T, R> function = (Function<T, R>) FUNCTION_MAP.get(functionName);
        if (function == null) {
            throw new IllegalArgumentException("No function registered with name: " + functionName);
        }
        T input = convertFromJsonStr(arguments, clazz);
        return function.apply(input);
    }
}

The caller's code is as follows:

java 复制代码
Function function = chatChoice.getMessage().getToolCalls().get(0).getFunction();

// execute function
WeatherResponse weatherResponse = FunctionExecutor.execute(function.getName(), Weather.class, function.getArguments());
System.out.println(weatherResponse)

The final result is:

shell 复制代码
WeatherResponse(location=Boston, MA, unit=null, temperature=2, description=sunny)
相关推荐
小小马喽_Thendras3 分钟前
ScheduledExecutorService 和Timer的区别
java·开发语言
小江的记录本10 分钟前
【Swagger】Swagger系统性知识体系全方位结构化总结
java·前端·后端·python·mysql·spring·docker
空太Jun11 分钟前
Spring Security 自定义数据库认证(初尝试)
java·数据库·spring
sinat_2554878118 分钟前
泛型·学习笔记
java·jvm·数据库·windows·python
QuZero23 分钟前
Java Synchronized principle
java·开发语言
明灯伴古佛27 分钟前
面试:Java中乐观锁的实现原理是什么
java·面试·职场和发展
SimonKing31 分钟前
白嫖党狂喜!魔塔社区每天2000次免费大模型调用,真香!
java·后端·程序员
lifallen1 小时前
Flink Agent 与 Checkpoint:主循环闭环与 Mailbox 事件驱动模型
java·大数据·人工智能·python·语言模型·flink
小则又沐风a1 小时前
类和对象----最终篇
java·前端·数据库
喵叔哟1 小时前
4.【.NET10 实战--孢子记账--产品智能化】--C# 14 新语法特性详解与实战应用
java·c#·.net