blame | history | patch | commit | commitdiff | ignore whitespace
Administrator
2025-09-02 f8f1d6a91b32077c69dd97334559fe3165ff79b6 
 src/main/java/cc/mrbird/febs/ai/strategy/Impl/HsLlmStrategyServiceImpl.java


@@ -1,29 +1,24 @@


package cc.mrbird.febs.ai.strategy.Impl;




import cc.mrbird.febs.ai.entity.AiTalkItem;


import cc.mrbird.febs.ai.res.ai.Report;


import cc.mrbird.febs.ai.res.memberTalk.ApiMemberTalkStreamVo;


import cc.mrbird.febs.ai.strategy.LlmStrategyService;


import cc.mrbird.febs.ai.strategy.enumerates.LlmStrategyContextEnum;


import cc.mrbird.febs.ai.strategy.param.LlmStrategyDto;


import cc.mrbird.febs.common.entity.FebsResponse;


import cc.mrbird.febs.common.exception.FebsException;


import cn.hutool.core.collection.CollUtil;


import cn.hutool.core.util.ObjectUtil;


import cn.hutool.core.util.StrUtil;


import com.alibaba.dashscope.common.Message;


import com.alibaba.dashscope.common.Role;


import com.fasterxml.jackson.core.JsonProcessingException;


import com.fasterxml.jackson.databind.JsonNode;


import com.volcengine.ark.runtime.model.completion.chat.*;


import com.volcengine.ark.runtime.service.ArkService;


import okhttp3.ConnectionPool;


import okhttp3.Dispatcher;


import org.springframework.stereotype.Component;


import org.springframework.util.StringUtils;


import reactor.core.publisher.Flux;




import javax.annotation.PostConstruct;


import javax.annotation.PreDestroy;


import java.util.ArrayList;


import java.util.HashMap;


import java.util.List;


import java.util.concurrent.TimeUnit;


import java.util.stream.Collectors;


@@ -33,7 +28,14 @@




    private ArkService service;




    private static final String ak = "AKLTZTQxZjMyZTUxMWJmNDEyNDkzNWExOGQ3ODllNzhhNmQ";


    private static final String sk = "TmpFeE1qZ3haREExTW1JeE5HRTBZVGc1WlRRNVlqWXpORGd5TWpsak5HWQ==";


    private static final String baseUrl = "https://ark.cn-beijing.volces.com/api/v3";


    private static final String LinkId = "ep-20250805124033-lhxbf";


    private static final Double temperature = 0.7;


    private static final Double topP = 0.9;


    private static final Integer maxTokens = 2048;


    private static final Double frequencyPenalty = 0.0;




    @PostConstruct


    public void init() {


@@ -47,9 +49,9 @@


        this.service = ArkService.builder()


                .dispatcher(dispatcher)


                .connectionPool(connectionPool)


                .baseUrl("https://ark.cn-beijing.volces.com/api/v3")


                .ak("AKLTZTQxZjMyZTUxMWJmNDEyNDkzNWExOGQ3ODllNzhhNmQ")


                .sk("TmpFeE1qZ3haREExTW1JeE5HRTBZVGc1WlRRNVlqWXpORGd5TWpsak5HWQ")


                .baseUrl(baseUrl)


                .ak(ak)


                .sk(sk)


                .build();


    }




@@ -96,10 +98,10 @@


                    .model(LinkId)


                    .messages(messages)


                    .stream(false)


                    .temperature(0.7) // 降低温度参数，提高确定性，可能提升速度


                    .topP(0.9)        // 调整topP参数


                    .maxTokens(2048)  // 减少最大token数


                    .frequencyPenalty(0.0)


                    .temperature(temperature) // 降低温度参数，提高确定性，可能提升速度


                    .topP(topP)        // 调整topP参数


                    .maxTokens(maxTokens)  // 减少最大token数


                    .frequencyPenalty(frequencyPenalty)


                    .build();




            List<ChatCompletionChoice> choices = service.createChatCompletion(chatCompletionRequest).getChoices();


@@ -115,7 +117,7 @@


    }




    @Override


    public Flux<FebsResponse> llmInvokeStreaming(List<LlmStrategyDto> dto) {


    public Flux<FebsResponse> llmInvokeStreamingWithThink(List<LlmStrategyDto> dto) {


        if (CollUtil.isEmpty(dto)){


            throw new FebsException("火山大模型初始化异常");


        }


@@ -126,10 +128,10 @@


                .messages(messages)


                .stream(true)


                .thinking(new ChatCompletionRequest.ChatCompletionRequestThinking("enabled"))


                .temperature(0.7)


                .topP(0.9)


                .maxTokens(2048)


                .frequencyPenalty(0.0)


                .temperature(temperature) // 降低温度参数，提高确定性，可能提升速度


                .topP(topP)        // 调整topP参数


                .maxTokens(maxTokens)  // 减少最大token数


                .frequencyPenalty(frequencyPenalty)


                .build();




        return Flux.from(service.streamChatCompletion(chatCompletionRequest))


@@ -144,21 +146,56 @@


                    }




                    ChatMessage message = choice.getMessage();


                    ApiMemberTalkStreamVo apiMemberTalkStreamVo = new ApiMemberTalkStreamVo();




                    // 处理 reasoning content


                    String reasoningContent = message.getReasoningContent();


                    if (StrUtil.isNotEmpty(reasoningContent)) {


                        apiMemberTalkStreamVo.setReasoningContent(reasoningContent);




                    HashMap<String, String> stringStringHashMap = new HashMap<>();


                    if (ObjectUtil.isNotEmpty(message.getReasoningContent())) {


                        stringStringHashMap.put(LlmStrategyContextEnum.THINK.name(),message.getReasoningContent().toString());


                    }


                    if (ObjectUtil.isNotEmpty(message.getContent())) {


                        stringStringHashMap.put(LlmStrategyContextEnum.CONTENT.name(),message.getContent().toString());


                    }


                    return new FebsResponse().success().data(stringStringHashMap);


                })


                .onErrorResume(throwable -> {


                    throw new FebsException(StrUtil.format("火山大模型流式调用AI服务失:{}",throwable));


                });


    }




    @Override


    public Flux<FebsResponse> llmInvokeStreamingNoThink(List<LlmStrategyDto> dto) {


        if (CollUtil.isEmpty(dto)){


            throw new FebsException("火山大模型初始化异常");


        }


        List<ChatMessage> messages = getMessages(dto);




        ChatCompletionRequest chatCompletionRequest = ChatCompletionRequest.builder()


                .model(LinkId)


                .messages(messages)


                .stream(true)


                .temperature(temperature) // 降低温度参数，提高确定性，可能提升速度


                .topP(topP)        // 调整topP参数


                .maxTokens(maxTokens)  // 减少最大token数


                .frequencyPenalty(frequencyPenalty)


                .build();




        return Flux.from(service.streamChatCompletion(chatCompletionRequest))


                .map(response -> {


                    if (response == null || response.getChoices() == null || response.getChoices().isEmpty()) {


                        return new FebsResponse().success().data("未响应，请重试");


                    }




                    // 安全处理 content


                    String content = "";


                    if (message.getContent() != null) {


                        content = message.getContent().toString();


                    ChatCompletionChoice choice = response.getChoices().get(0);


                    if (choice == null || choice.getMessage() == null) {


                        return new FebsResponse().success().data("END");


                    }


                    apiMemberTalkStreamVo.setContent(content);


                    return new FebsResponse().success().data(apiMemberTalkStreamVo);


                    ChatMessage message = choice.getMessage();




                    HashMap<String, String> stringStringHashMap = new HashMap<>();


                    if (ObjectUtil.isNotEmpty(message.getContent())) {


                        stringStringHashMap.put(LlmStrategyContextEnum.CONTENT.name(),message.getContent().toString());


                    }


                    return new FebsResponse().success().data(stringStringHashMap);


                })


                .onErrorResume(throwable -> {


                    throw new FebsException(StrUtil.format("火山大模型流式调用AI服务失:{}",throwable));