linux-china / chatgpt-spring-boot-starter
- среда, 21 июня 2023 г. в 00:00:06
Spring Boot ChatGPT Starter
Spring Boot ChatGPT starter with ChatGPT chat and functions support.
Add chatgpt-spring-boot-starter
dependency in your pom.xml.
<dependency>
<groupId>org.mvnsearch</groupId>
<artifactId>chatgpt-spring-boot-starter</artifactId>
<version>0.2.1</version>
</dependency>
Add openai.api.key
in application.properties
:
# OpenAI API Token, or you can set environment variable OPENAI_API_KEY
openai.api.key=sk-xxxx
If you want to use Azure OpenAI, you can add openai.api.url
in application.properties
:
openai.api.key=1138xxxx9037
openai.api.url=https://YOUR_RESOURCE_NAME.openai.azure.com/openai/deployments/YOUR_DEPLOYMENT_NAME/chat/completions?api-version=2023-05-15
@RestController
public class ChatRobotController {
@Autowired
private ChatGPTService chatGPTService;
@PostMapping("/chat")
public Mono<String> chat(@RequestBody String content) {
return chatGPTService.chat(ChatCompletionRequest.of(content))
.map(ChatCompletionResponse::getReplyText);
}
@GetMapping("/stream-chat")
public Flux<String> streamChat(@RequestParam String content) {
return chatGPTService.stream(ChatCompletionRequest.of(content))
.map(ChatCompletionResponse::getReplyText);
}
}
@Component
and implement GPTFunctionsStub
interface. Annotate GPT functions
with @GPTFunction
annotation, and annotate function parameters with @Parameter
annotation. @Nonnull
means that
the parameter is required.import jakarta.annotation.Nonnull;
@Component
public class GPTFunctions implements GPTFunctionsStub {
public record SendEmailRequest(
@Nonnull @Parameter("Recipients of email") List<String> recipients,
@Nonnull @Parameter("Subject of email") String subject,
@Parameter("Content of email") String content) {
}
@GPTFunction(name = "send_email", value = "Send email to receiver")
public String sendEmail(SendEmailRequest request) {
System.out.println("Recipients: " + String.join(",", request.recipients));
System.out.println("Subject: " + request.subject);
System.out.println("Content:\n" + request.content);
return "Email sent to " + String.join(",", request.recipients) + " successfully!";
}
public record SQLQueryRequest(
@Parameter(required = true, value = "SQL to query") String sql) {
}
@GPTFunction(name = "execute_sql_query", value = "Execute SQL query and return the result set")
public String executeSQLQuery(SQLQueryRequest request) {
System.out.println("Execute SQL: " + request.sql);
return "id, name, salary\n1,Jackie,8000\n2,Libing,78000\n3,Sam,7500";
}
}
response.getReplyCombinedText()
or chatMessage.getFunctionCall().getFunctionStub().call()
:public class ChatGPTServiceImplTest {
@Test
public void testChatWithFunctions() throws Exception {
final String prompt = "Hi Jackie, could you write an email to Libing(libing.chen@gmail.com) and Sam(linux_china@hotmail.com) and invite them to join Mike's birthday party at 4 pm tomorrow? Thanks!";
final ChatCompletionRequest request = ChatCompletionRequest.functions(prompt, List.of("send_email"));
final ChatCompletionResponse response = chatGPTService.chat(request).block();
// display reply combined text with function call
System.out.println(response.getReplyCombinedText());
// call function manually
for (ChatMessage chatMessage : response.getReply()) {
final FunctionCall functionCall = chatMessage.getFunctionCall();
if (functionCall != null) {
final Object result = functionCall.getFunctionStub().call();
System.out.println(result);
}
}
}
@Test
public void testExecuteSQLQuery() {
String context = "You are SQL developer. Write SQL according to requirements, and execute it in MySQL database.";
final String prompt = "Query all employees whose salary is greater than the average.";
final ChatCompletionRequest request = ChatCompletionRequest.functions(prompt, List.of("execute_sql_query"));
// add prompt context as system message
request.addMessage(ChatMessage.systemMessage(context));
final ChatCompletionResponse response = chatGPTService.chat(request).block();
System.out.println(response.getReplyCombinedText());
}
}
If you want to have a simple test for ChatGPT functions, you can install ChatGPT with Markdown JetBrains IDE Plugin, and take a look at chat.gpt file.
Please refer OpenAIProxyController.
@RestController
public class OpenAIProxyController {
@Autowired
private OpenAIChatAPI openAIChatAPI;
@PostMapping("/v1/chat/completions")
public Publisher<ChatCompletionResponse> completions(@RequestBody ChatCompletionRequest request) {
return openAIChatAPI.proxy(request);
}
}
Of course, you can use standard URL http://localhost:8080/v1/chat/completions
to call Azure OpenAI API.
How to manage prompts in Java? Now my suggestion is to adopt properties file format, and use MessageFormat to format. Please take a look at PromptManager