springboot 对接豆包大模型知识库 java SDK

前提

对于需要对接豆包大模型知识库能力的个人用户或企业，可以参考该SDK，从官方获取的。包含流式和非流式请求 demo，基于知识库检索生成大模型对话能力！本人已对接实现验证，有疑问可以私聊！

代码如下

引入maven依赖

        <!-- 豆包 -->
        <dependency>
            <groupId>com.volcengine</groupId>
            <artifactId>volcengine-java-sdk-ark-runtime</artifactId>
            <version>LATEST</version>
        </dependency>
        <dependency>
            <groupId>com.volcengine</groupId>
            <artifactId>volc-sdk-java</artifactId>
            <version>1.0.206</version>
        </dependency>

java代码

package org.example;

import com.fasterxml.jackson.annotation.JsonProperty;
import com.fasterxml.jackson.core.type.TypeReference;
import com.fasterxml.jackson.databind.ObjectMapper;
import com.volcengine.auth.ISignerV4;
import com.volcengine.auth.impl.SignerV4Impl;
import com.volcengine.model.Credentials;
import com.volcengine.service.SignableRequest;
import lombok.Data;
import lombok.experimental.Accessors;
import org.apache.http.Header;
import org.apache.http.HttpResponse;
import org.apache.http.NameValuePair;
import org.apache.http.client.HttpClient;
import org.apache.http.client.config.RequestConfig;
import org.apache.http.client.methods.HttpPost;
import org.apache.http.client.utils.URIBuilder;
import org.apache.http.entity.StringEntity;
import org.apache.http.impl.client.HttpClients;
import org.apache.http.util.EntityUtils;

import java.io.BufferedReader;
import java.io.InputStream;
import java.io.InputStreamReader;
import java.net.URI;
import java.util.*;

public class Main {
    public static final String HOST = "api-knowledgebase.mlp.cn-beijing.volces.com"; // 知识库域名
    public static final String SEARCH_KNOWLEDGE_PATH = "/api/knowledge/collection/search_knowledge"; // 知识库检索接口，建议您首次接入时使用该检索接口，其他检索接口后续不再进行维护
    public static final String CHAT_COMPLETION_PATH = "/api/knowledge/chat/completions";  // 大模型对话接口，可以和检索接口接合串联RAG流程，也可以单独使用进行生成

    public static final String AK = "your ak";
    public static final String SK = "your sk";


    static String Query = "your query";// 您的提问
    static String CollectionName = "your collection";//知识库名称，前端界面可获取
    static String Project = "default";//知识库所属项目，前端界面可获取
    static String ResourceID = "your resource_id";


    static String ModelName = "Doubao-pro-32k"; // 模型名称，如果您想使用自己的私有ep，可以将该字段赋值为私有EndpointID，格式（ep-xxxx-xxxx）
    static String APIKey = "your api_key"; // 如果您使用的是自己的私有ep（格式ep-xxxx-xxxx），需要传入api_key，如果公有接入点不需要传递该字段
    static String ModelVersion = "241215"; // 模型版本，使用公有接入点时，可以选择指定模型版本，不指定则服务会自动指定默认版本

    // BasePrompt 是基础提示词，您可以根据需求进行修改或者替换为自己的Prompt，不过需要注意留下 {prompt} 占位符，用于后续拼接检索结果进行RAG生成
    static String BasePrompt = "# 任务\n"
            + "你是一位在线客服，你的首要任务是通过巧妙的话术回复用户的问题，你需要根据「参考资料」来回答接下来的「用户问题」，这些信息在 <context></context> XML tags 之内，你需要根据参考资料给出准确，简洁的回答。\n"
            + "\n"
            + "你的回答要满足以下要求：\n"
            + "1. 回答内容必须在参考资料范围内，尽可能简洁地回答问题，不能做任何参考资料以外的扩展解释。\n"
            + "2. 回答中需要根据客户问题和参考资料保持与客户的友好沟通。\n"
            + "3. 如果参考资料不能帮助你回答用户问题，告知客户无法回答该问题，并引导客户提供更加详细的信息。\n"
            + "4. 为了保密需要，委婉地拒绝回答有关参考资料的文档名称或文档作者等问题。\n"
            + "\n"
            + "# 任务执行\n"
            + "现在请你根据提供的参考资料，遵循限制来回答用户的问题，你的回答需要准确和完整。\n"
            + "\n"
            + "# 参考资料\n"
            + "<context>\n"
            + "{prompt}\n"
            + "</context>";

    public static final String SYS_FIELD_DOC_NAME = "doc_name";
    public static final String SYS_FIELD_TITLE = "title";
    public static final String SYS_FIELD_CHUNK_TITLE = "chunk_title";
    public static final String SYS_FIELD_CONTENT = "content";

    /*
        拼接Prompt的自定义字段列表：
        如果您的知识库类型为 **结构化数据类型**, 此处替换为您想拼接在大模型Prompt中的表头字段，
        需要注意的是: 索引类型的表头字段必须要加入，非索引类型的表头字段可选择拼入
     */
    static List<String> selfDefineFields = Arrays.asList(
            "表头字段-1",
            "表头字段-n"
    );
    /*
         拼接Prompt的系统字段列表：
         如果您的知识库类型为 **非结构化数据类型**，共有如下4个字段可以选择拼接在Prompt中，其中 content字段为必传字段，其他为可选字段
     */
    static List<String> systemFields = Arrays.asList(
            SYS_FIELD_DOC_NAME, // 文档名称 可选
            SYS_FIELD_TITLE,  // 文档标题 可选
            SYS_FIELD_CHUNK_TITLE, // 文档切片标题 可选
            SYS_FIELD_CONTENT  // 文档切片内容 必传
    );

    public static final PromptExtraContext promptExtraContextExample = new PromptExtraContext(selfDefineFields, systemFields);

    @Data
    public static class PromptExtraContext {
        private List<String> SelfDefineFields;
        private List<String> SystemFields;

        public PromptExtraContext(List<String> selfDefineFields, List<String> systemFields) {
            this.SelfDefineFields = selfDefineFields;
            this.SystemFields = systemFields;
        }
    }


    @Data
    public static class SearchKnowledgeRequest {
        private String name;
        private String project;
        @JsonProperty("resource_id")
        private String resourceId;
        private String query;
        private Integer limit;
        @JsonProperty("dense_weight")
        private Float denseWeight;
        @JsonProperty("md_search")
        private Boolean mdSearch;
        @JsonProperty("query_param")
        private QueryParam queryParam;
        @JsonProperty("pre_processing")
        private PreProcessing preProcessing;
        @JsonProperty("post_processing")
        private PostProcessing postProcessing;
    }

    @Data
    public static class QueryParam {
        @JsonProperty("doc_filter")
        private Object docFilter;
    }

    @Data
    public static class PreProcessing {
        @JsonProperty("need_instruction")
        private Boolean needInstruction;
        private Boolean rewrite;
        private List<MessageParam> messages;
        @JsonProperty("return_token_usage")
        private Boolean returnTokenUsage;
    }

    @Data
    public static class PostProcessing {
        @JsonProperty("rerank_switch")
        private Boolean rerankSwitch;

        @JsonProperty("rerank_model")
        private String rerankModel;
        @JsonProperty("rerank_only_chunk")
        private Boolean rerankOnlyChunk;
        @JsonProperty("retrieve_count")
        private Integer retrieveCount;
        @JsonProperty("endpoint_id")
        private String endpointId;
        @JsonProperty("chunk_diffusion_count")
        private Integer chunkDiffusionCount;
        @JsonProperty("chunk_group")
        private Boolean chunkGroup;
        @JsonProperty("chunk_score_aggr_type")
        private String chunkScoreAggrType;
        @JsonProperty("chunk_extra_content")
        private Map<String, Object> chunkExtraContent;
        @JsonProperty("get_attachment_link")
        private Boolean getAttachmentLink;
    }

    @Data
    public static class MessageParam {
        private String role;
        private Object content;

        MessageParam(String role, Object content) {
            this.role = role;
            this.content = content;
        }
    }

    @Data
    public static class ChatMessageImageURL {
        private String url;

        ChatMessageImageURL(String url) {
            this.url = url;
        }

    }

    public enum ChatCompletionMessageContentPartType {
        TEXT, IMAGE_URL
    }

    @Data
    public static class ChatCompletionMessageContentPart {
        private ChatCompletionMessageContentPartType type;
        private String text;
        private ChatMessageImageURL imageURL;

        public ChatCompletionMessageContentPart(ChatCompletionMessageContentPartType type, String text, ChatMessageImageURL imageURL) {
            this.type = type;
            this.text = text;
            this.imageURL = imageURL;
        }
    }

    @Data
    public static class BaseResponse<T> {
        private Integer code;
        private String message;
        @JsonProperty("request_id")
        private String requestId;
        private T data;
    }

    @Data
    public static class CollectionSearchKnowledgeResponseData {
        @JsonProperty("collection_name")
        private String collectionName; // 知识库collection_name
        private Integer count; // 返回切片数
        @JsonProperty("rewrite_query")
        private String rewriteQuery; // 改写后的query,改写功能开启时返回
        @JsonProperty("token_usage")
        private TotalTokenUsage tokenUsage; // 模型使用详情
        @JsonProperty("result_list")
        private List<CollectionSearchResponseItem> resultList; //返回切片信息
    }

    @Data
    public static class TotalTokenUsage {
        @JsonProperty("embedding_token_usage")
        private ModelTokenUsage embeddingUsage;
        @JsonProperty("rerank_token_usage")
        private Long rerankUsage;
        @JsonProperty("llm_token_usage")
        private ModelTokenUsage llmUsage;
        @JsonProperty("rewrite_token_usage")
        private ModelTokenUsage rewriteUsage;
    }

    /*
        检索返回的切片信息，具体字段
     */
    @Data
    public static class CollectionSearchResponseItem {
        private String id;
        private String content;
        @JsonProperty("md_content")
        private String mdContent;
        private Double score;
        @JsonProperty("point_id")
        private String pointId;
        @JsonProperty("origin_text")
        private String originText;
        @JsonProperty("original_question")
        private String originalQuestion;
        @JsonProperty("chunk_title")
        private String chunkTitle;
        @JsonProperty("chunk_id")
        private Integer chunkId;
        @JsonProperty("process_time")
        private Long processTime;
        @JsonProperty("rerank_score")
        private Double rerankScore;
        @JsonProperty("doc_info")
        private CollectionSearchResponseItemDocInfo docInfo;
        @JsonProperty("recall_position")
        private Integer recallPosition;
        @JsonProperty("rerank_position")
        private Integer rerankPosition;
        @JsonProperty("chunk_type")
        private String chunkType;
        @JsonProperty("chunk_source")
        private String chunkSource;
        @JsonProperty("update_time")
        private Long updateTime;
        @JsonProperty("chunk_attachment")
        private List<ChunkAttachment> chunkAttachmentList;
        @JsonProperty("table_chunk_fields")
        private List<PointTableChunkField> tableChunkFields;
        @JsonProperty("original_coordinate")
        private ChunkPositions originalCoordinate;
    }

    @Data
    public static class CollectionSearchResponseItemDocInfo {
        @JsonProperty("doc_id")
        private String docId;
        @JsonProperty("doc_name")
        private String docName;
        @JsonProperty("create_time")
        private Long createTime;
        @JsonProperty("doc_type")
        private String docType;
        @JsonProperty("doc_meta")
        private String docMeta;
        private String source;
        private String title;
    }

    @Data
    public static class ChunkAttachment {
        private String uuid;
        private String caption;
        private String type;
        private String link;
    }

    @Data
    public static class PointTableChunkField {
        @JsonProperty("field_name")
        private String fieldName;
        @JsonProperty("field_value")
        private Object fieldValue;  // Using Object to accommodate any type of field value
    }

    @Data
    public static class ChunkPositions {
        @JsonProperty("page_no")
        private List<Integer> pageNo;
        private List<List<Double>> bbox;  // A list of lists to represent bounding box coordinates
    }


    @Data
    public static class ChatCompletionRequest {
        private String model;
        @JsonProperty("model_version")
        private String modelVersion;
        private String project;
        private Boolean Stream;
        @JsonProperty("return_token_usage")
        private Boolean ReturnTokenUsage;
        @JsonProperty("api_key")
        private String APIKey;
        @JsonProperty("max_tokens")
        private Integer MaxTokens;
        private double Temperature;
        private List<MessageParam> messages;
    }


    @Data
    public static class CollectionChatCompletionResponseData {
        @JsonProperty("generated_answer")
        private String GenerateAnswer; // 模型生成文本
        private String Usage; // 该字段是ModelTokenUsage类型序列化而来，如果需要获取usage信息，需要自行进行反序列化
        @JsonProperty("reasoning_content")
        private String ReasoningContent; //模型推理过程内容，仅推理模型会有
        private Boolean End;//流式响应标记是否最后一个
    }

    @Data
    public static class ModelTokenUsage {
        @JsonProperty("prompt_tokens")
        private Integer PromptTokens;  // 请求文本的分词数
        @JsonProperty("completion_tokens")
        private Integer CompletionTokens; // 生成文本的分词数, LLM对话模型才有值, 其他模型都是0
        @JsonProperty("total_tokens")
        private Integer TotalTokens; // PromptTokens + CompletionTokens
    }


    public static class PromptResult {
        String prompt;
        List<String> imageURLs;
        Exception error;

        public PromptResult(String prompt, List<String> imageURLs, Exception error) {
            this.prompt = prompt;
            this.imageURLs = imageURLs;
            this.error = error;
        }
    }


    public static String toJson(Object obj) {
        try {
            // 创建 ObjectMapper 实例
            ObjectMapper objectMapper = new ObjectMapper();
            // 将对象转换为 JSON 字符串
            return objectMapper.writeValueAsString(obj);
        } catch (Exception e) {
            e.printStackTrace();
        }
        return null;
    }


    public static SignableRequest prepareRequest(String host, String path, String method, List<NameValuePair> params, String body, String ak, String sk) throws Exception {
        SignableRequest request = new SignableRequest();
        request.setMethod(method);
        request.setHeader("Accept", "application/json");
        request.setHeader("Content-Type", "application/json");
        request.setHeader("Host", HOST);
        request.setEntity(new StringEntity(body, "utf-8"));

        URIBuilder builder = request.getUriBuilder();
        builder.setScheme("https");
        builder.setHost(host);
        builder.setPath(path);
        if (params != null) {
            builder.setParameters(params);
        }

        RequestConfig requestConfig = RequestConfig.custom().setSocketTimeout(120000).setConnectTimeout(12000).build();
        request.setConfig(requestConfig);

        Credentials credentials = new Credentials("cn-north-1", "air");
        credentials.setAccessKeyID(ak);
        credentials.setSecretAccessKey(sk);

        // 签名
        ISignerV4 ISigner = new SignerV4Impl();
        ISigner.sign(request, credentials);

        return request;
    }


    /**
     * 知识库检索请求参数生成：以下详细展示了部分参数的传递规则，其余参数请参考官方接口文档，如果您想快速接入进行测试，可以只传入参数集合的最小集，其余参数可以使用默认值。
     * 必传参数如下：
     * 1. resourceId (也可使用resource_id 或 name + project, 二选一)
     * 2. query：用户问题
     */
    public static SearchKnowledgeRequest GenerateSearchKnowledgeRequest() {
        SearchKnowledgeRequest requestParams = new SearchKnowledgeRequest();
        requestParams.setQuery(Query);
        requestParams.setName(CollectionName); // 可选
        requestParams.setProject(Project); // 可选
        //requestParams.setResourceId(ResourceID);// 知识库resource_id (二选一，查询时，可使用resource_id 或 name + project)
        requestParams.setLimit(10);// 检索最终返回的结果数量, 不传递默认返回10条
        requestParams.setDenseWeight(0.5f);//

        /*
            检索预处理结构体，提供改写和意图识别等附加功能，可按需开启，非必传
         */
        PreProcessing preprocessing = new PreProcessing();
        preprocessing.setNeedInstruction(true);//设置向量化模型是否拼接指令，查看官方文档
        preprocessing.setReturnTokenUsage(true);// 检索是否返回token使用信息
        preprocessing.setRewrite(false); // 问题改写开关，检索时提供根据多轮对话进行改写功能,默认不开启

        requestParams.setPreProcessing(preprocessing);

        /*
            检索后处理结构体，提供Rerank等功能，可按需开启，非必传
         */
        PostProcessing postProcessing = new PostProcessing();
        postProcessing.setRerankSwitch(false);// 重排开关，默认不开启
        postProcessing.setRetrieveCount(25);// 进入重排的切片数量，重排打开时生效，需要大于limit,当limit=10，默认值为25,可按需调整
        postProcessing.setChunkGroup(true);// 是否对召回切片按照文档进行聚合

        return requestParams;
    }

    // 检索接口调用
    public static BaseResponse<CollectionSearchKnowledgeResponseData> SearchKnowledge() throws Exception {
        SearchKnowledgeRequest searchRequestParams = GenerateSearchKnowledgeRequest();
        String searchRequestParamsJson = toJson(searchRequestParams);
        //可自行处理做更加详细的错误处理
        try {
            SignableRequest signableRequest = prepareRequest(HOST, SEARCH_KNOWLEDGE_PATH, "POST", null, searchRequestParamsJson, AK, SK);
            URI uri = new URIBuilder()
                    .setScheme("https")
                    .setHost(HOST)
                    .setPath(SEARCH_KNOWLEDGE_PATH)
                    .build();

            HttpPost httpPost = new HttpPost(uri);
            httpPost.setConfig(signableRequest.getConfig());
            httpPost.setEntity(signableRequest.getEntity());
            for (Header header : signableRequest.getAllHeaders()) {
                httpPost.setHeader(header.getName(), header.getValue());
            }

            HttpClient httpClient = HttpClients.createDefault();

            HttpResponse response = httpClient.execute(httpPost);

            int statusCode = response.getStatusLine().getStatusCode();
            String responseBody = EntityUtils.toString(response.getEntity());

            BaseResponse<CollectionSearchKnowledgeResponseData> resp = new ObjectMapper().readValue(responseBody, new TypeReference<BaseResponse<CollectionSearchKnowledgeResponseData>>() {
            });
            return resp;
        } catch (Exception e) {
            e.printStackTrace();
            throw e;
        }
    }


    public static ChatCompletionRequest GenerateChatCompletionRequest(Boolean Stream, List<MessageParam> messages) {
        ChatCompletionRequest requestParams = new ChatCompletionRequest();
        requestParams.setModel(ModelName);// 传入模型名称即代表使用公有接入点, 如果使用私有ep，此处替换为私有ep即可，格式 ep-xxx-xxx
        requestParams.setModelVersion(ModelVersion);// 模型版本，使用公有接入点时，可以选择指定模型版本，不指定则服务会自动指定默认版本
        requestParams.setReturnTokenUsage(true); //是否返回LLM调用的token使用情况
        requestParams.setAPIKey(APIKey); // 使用私有ep(即model为ep-xxx-xxx)时，必须传递此参数才能生效,
        requestParams.setMaxTokens(4096); // 输出最大token数量，目前最大支持4096
        requestParams.setTemperature(0.7f); // 模型温度,取值范围0~1，值越大随机性越大
        requestParams.setStream(Stream); // 模型结果是否流式返回
        requestParams.setMessages(messages); // 模型对话信息，需要保证拼接格式,对话顺序正确，详细请查看官方文档

        return requestParams;
    }


    public static String GetContentForPrompt(CollectionSearchResponseItem item, int imageNum) {
        String content = item.content;

        if (!Objects.equals(item.originalQuestion, "")) {
            return String.format("当询问到相似问题时，请参考对应答案进行回答：问题：“%s”。答案：“%s”", item.getOriginalQuestion(), content);
        }
        if (imageNum > 0 && item.getChunkAttachmentList() != null && !item.getChunkAttachmentList().isEmpty()
                && item.getChunkAttachmentList().get(0).getLink() != null && !item.getChunkAttachmentList().get(0).getLink().isEmpty()) {
            String placeholder = String.format("<img>图片%d</img>", imageNum);
            return content + placeholder;
        }
        return content;
    }


    // 根据检索结果和提供的BasePrompt拼接生成用于调用大模型的SysPrompt
    public static PromptResult GeneratePrompt(BaseResponse<CollectionSearchKnowledgeResponseData> resp) {
        if (resp == null) {
            return new PromptResult("", null, new IllegalArgumentException("response is nil"));
        }
        if (resp.getCode() != 0) {
            return new PromptResult("", null, new IllegalArgumentException(resp.getMessage()));
        }

        StringBuilder promptBuilder = new StringBuilder();
        List<String> imageURLs = new ArrayList<>();
        boolean usingVLM = isVisionModel("ModelName");
        int imageCnt = 0;

        for (CollectionSearchResponseItem item : resp.data.resultList) {
            if (usingVLM && item.getChunkAttachmentList() != null && !item.getChunkAttachmentList().isEmpty()) {
                String link = item.getChunkAttachmentList().get(0).getLink();
                if (link != null && !link.isEmpty()) {
                    imageURLs.add(link);
                    imageCnt++;
                }
            }

            CollectionSearchResponseItemDocInfo docInfo = item.getDocInfo();

            // 拼接用户指定的系统字段
            for (String sysField : promptExtraContextExample.SystemFields) {
                switch (sysField) {
                    case SYS_FIELD_DOC_NAME:
                        promptBuilder.append(String.format("%s: %s\n", sysField, docInfo.getDocName()));
                        break;
                    case SYS_FIELD_TITLE:
                        promptBuilder.append(String.format("%s: %s\n", sysField, docInfo.getTitle()));
                        break;
                    case SYS_FIELD_CHUNK_TITLE:
                        promptBuilder.append(String.format("%s: %s\n", sysField, item.getChunkTitle()));
                        break;
                    case SYS_FIELD_CONTENT:
                        promptBuilder.append(String.format("%s: %s\n", sysField, GetContentForPrompt(item, imageCnt)));
                        break;
                }
            }

            // 注：针对知识库为结构化类型- prompt拼接用户指定的自定义字段（非结构化的知识库不用处理）
            for (String selfField : promptExtraContextExample.SelfDefineFields) {
                if (item.getTableChunkFields() != null && !item.getTableChunkFields().isEmpty()) {
                    for (PointTableChunkField tableChunkField : item.getTableChunkFields()) {
                        if (tableChunkField.getFieldName().equals(selfField)) {
                            promptBuilder.append(String.format("%s: %s\n", tableChunkField.getFieldName(), tableChunkField.getFieldValue()));
                        }
                    }
                }
            }

            promptBuilder.append("---\n");
        }

        String finalPrompt = BasePrompt.replace("{prompt}", promptBuilder);
        //System.out.println(finalPrompt);
        return new PromptResult(finalPrompt, imageURLs, null);
    }


    /**
     * 检查模型名称是否包含 "vision" 字符串，以判断是否是视觉模型。
     *
     * @param modelName 模型名称
     * @return 如果模型名称包含 "vision"，则返回 true，否则返回 false。
     */
    public static boolean isVisionModel(String modelName) {
        return modelName != null && modelName.contains("vision");
    }

    public static BaseResponse<CollectionChatCompletionResponseData> ChatCompletion(List<MessageParam> messages) throws Exception {
        ChatCompletionRequest chatCompletionRequest = GenerateChatCompletionRequest(false, messages);
        String chatCompletionRequestParamsJson = toJson(chatCompletionRequest);
        try {
            SignableRequest signableRequest = prepareRequest(HOST, CHAT_COMPLETION_PATH, "POST", null, chatCompletionRequestParamsJson, AK, SK);
            URI uri = new URIBuilder()
                    .setScheme("https")
                    .setHost(HOST)
                    .setPath(CHAT_COMPLETION_PATH)
                    .build();

            HttpPost httpPost = new HttpPost(uri);
            httpPost.setConfig(signableRequest.getConfig());
            httpPost.setEntity(signableRequest.getEntity());
            for (Header header : signableRequest.getAllHeaders()) {
                httpPost.setHeader(header.getName(), header.getValue());
            }

            HttpClient httpClient = HttpClients.createDefault();
            HttpResponse response = httpClient.execute(httpPost);

            int statusCode = response.getStatusLine().getStatusCode();
            String responseBody = EntityUtils.toString(response.getEntity());

            BaseResponse<CollectionChatCompletionResponseData> resp = new ObjectMapper().readValue(responseBody, new TypeReference<BaseResponse<CollectionChatCompletionResponseData>>() {
            });

            return resp;
        } catch (Exception e) {
            e.printStackTrace();
            throw e;
        }
    }


    /**
     * 大模型流式调用 Chat-Completion，流式返回生成内容和其他所有信息
     */
    public static void ChatCompletionStream(List<MessageParam> messages) throws Exception {
        ChatCompletionRequest chatCompletionRequest = GenerateChatCompletionRequest(true, messages);
        // 序列化
        String chatCompletionRequestParamsJson = toJson(chatCompletionRequest);
        try {
            SignableRequest signableRequest = prepareRequest(HOST, CHAT_COMPLETION_PATH, "POST", null, chatCompletionRequestParamsJson, AK, SK);
            URI uri = new URIBuilder()
                    .setScheme("https")
                    .setHost(HOST)
                    .setPath(CHAT_COMPLETION_PATH)
                    .build();

            HttpPost httpPost = new HttpPost(uri);
            httpPost.setConfig(signableRequest.getConfig());
            httpPost.setEntity(signableRequest.getEntity());
            for (Header header : signableRequest.getAllHeaders()) {
                httpPost.setHeader(header.getName(), header.getValue());
            }
            httpPost.setHeader("Accept", "text/event-stream"); // 设置Accept头为text/event-stream

            HttpClient httpClient = HttpClients.createDefault();
            HttpResponse response = httpClient.execute(httpPost);

            InputStream stream = response.getEntity().getContent();
            BufferedReader reader = new BufferedReader(new InputStreamReader(stream));
            StringBuilder answer = new StringBuilder();
            String usage = "";
            String line;
            while ((line = reader.readLine()) != null) {
                if (line.startsWith("data:") && line.length() > 5) {
                    String content = line.substring(5);
                    BaseResponse<CollectionChatCompletionResponseData> resp = new ObjectMapper().readValue(content, new TypeReference<BaseResponse<CollectionChatCompletionResponseData>>() {
                    });
                    System.out.println(resp.data.GenerateAnswer);
                    answer.append(resp.data.GenerateAnswer);
                    // 最后一个stream中包含usage信息
                    if (resp.data.getEnd() != null && resp.data.End) {
                        usage = resp.data.Usage;
                        //System.out.println(resp.data.Usage);
                    }
                }
            }

            System.out.printf("大模型流式调用返回结果: %s\n", answer);
            System.out.printf("大模型流式调用返回token使用情况: %s\n", usage);
        } catch (Exception e) {
            e.printStackTrace();
            throw e;
        }
    }

    public static List<MessageParam> BuildMessages(String prompt, String query, List<String> images) {
        List<MessageParam> messages = new ArrayList<>();

        if (!images.isEmpty()) {
            /*
                如果采用的是VLM的模型，使用该分支拼接生成message, 拼接的message的role顺序如下：
                 [system, user, assistant, user, assistant, user，assistant, user...]
             */
            List<ChatCompletionMessageContentPart> multiModalMessage = new ArrayList<>();
            multiModalMessage.add(new ChatCompletionMessageContentPart(ChatCompletionMessageContentPartType.TEXT, query, null));

            for (String imageURL : images) {
                multiModalMessage.add(new ChatCompletionMessageContentPart(ChatCompletionMessageContentPartType.IMAGE_URL, null, new ChatMessageImageURL(imageURL)));
            }

            messages.add(new MessageParam("system", prompt));
            messages.add(new MessageParam("user", multiModalMessage));
        } else {
            /*
                如果使用的是普通的文本LLM模型，使用该分支拼接生成message，拼接的多轮对话message的role顺序如下：
                 [system, user, assistant, user, assistant, user，assistant, user...]
             */
            messages.add(new MessageParam("system", prompt)); // 系统Prompt
            messages.add(new MessageParam("user", query)); // 用户首个提问
            //messages.add(new MessageParam("assistant", new ChatCompletionMessageContent(query, null))); // LLM回答
            //messages.add(new MessageParam("user", new ChatCompletionMessageContent(query, null))); // 用户第二轮提问
            //messages.add(new MessageParam("assistant", new ChatCompletionMessageContent(query, null))); // LLM第二轮回答
            //...
        }
        //System.out.println("messages: " + messages);
        return messages;
    }


    public static void RAG(Boolean stream) throws Exception {
        try {
            // 1.先进行知识库检索
            BaseResponse<CollectionSearchKnowledgeResponseData> searchResponse = SearchKnowledge();

            // 2.生成大模型提示词Prompt
            PromptResult promptResult = GeneratePrompt(searchResponse);

            // 3.拼接message对话信息
            List<MessageParam> messages = BuildMessages(promptResult.prompt, Query, promptResult.imageURLs);

            // 4.调用大模型LLM进行生成（流式/非流式）
            if (stream) {
                // 流式调用
                ChatCompletionStream(messages);
            } else {
                // 非流式调用
                BaseResponse<CollectionChatCompletionResponseData> chatResponse = ChatCompletion(messages);
                String generatedAnswer = chatResponse.data.GenerateAnswer;
                //System.out.printf("大模型非流式调用返回生成结果: %s\n",generatedAnswer);
                String usage = chatResponse.data.Usage; // 字符串，可自行进行反序列化解析
                //System.out.printf("大模型流式调用返回token使用情况: %s\n",usage);
            }
        } catch (Exception e) {
            throw new RuntimeException(e);
        }
    }

    // 3种流程，可以按照自身需求使用
    public static void main(String[] args) throws Exception {
        // 1.只使用知识库检索
        //BaseResponse<CollectionSearchKnowledgeResponseData> searchResponse = SearchKnowledge();
        //2. RAG检索+生成流程-非流式
        //RAG(false);
        //3. RAG检索+生成流程-流式
        RAG(true);
    }
}