manager/my-worker
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

1. 主动停止的响应的消息仍然存储对话数据；2. 对话消息框当手动向上滚动的时候，停止滚动到底部，方便阅读已生成的消息；3. 消息内容，对 JSON 格式的数据，进行格式化显示；4. 支持“深度思考”按钮，可展开和折叠深度思考的消息；对 AI 的用户消息进行转义，避免不会支持的消息报错；5. 当 AI 接口调用异常的时候，给于用户提示实际的接口返回内容；6. 新增自动更新对话标题；6. 加载消息过程中，避免再次发送新消息和切换对话。

Browse Source
This commit is contained in:
thinkgem
2025-04-21 13:24:23 +08:00
parent 695762b34c
commit 1b8b6162f4
6 changed files with 272 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
modules/cms-ai/src/main/java/com/jeesite/modules/cms/ai/config/CmsAiChatConfig.java
View File

@@ -5,9 +5,11 @@
package com.jeesite.modules.cms.ai.config;
import com.jeesite.common.datasource.DataSourceHolder;
import com.jeesite.modules.cms.ai.properties.CmsAiProperties;
import com.jeesite.modules.cms.ai.tools.CmsAiTools;
import org.springframework.ai.chat.client.ChatClient;
import org.springframework.boot.autoconfigure.condition.ConditionalOnProperty;
import org.springframework.boot.context.properties.EnableConfigurationProperties;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;
import org.springframework.context.annotation.Primary;
@@ -18,6 +20,7 @@ import org.springframework.jdbc.core.JdbcTemplate;
* @author ThinkGem
*/
@Configuration
@EnableConfigurationProperties(CmsAiProperties.class)
public class CmsAiChatConfig {
/**
@@ -25,19 +28,21 @@ public class CmsAiChatConfig {
* @author ThinkGem
*/
@Bean
public ChatClient chatClient(ChatClient.Builder builder) {
return builder
.defaultSystem("""
## 人物设定
你是我的知识库AI助手，你把我当作朋友，耐心真诚地回复我提出的相关问题。
你需要遵循以下原则，与关注者进行友善而有价值的沟通。
## 表达方式：
1. 使用简体中文回答我的问题。
2. 使用幽默有趣的方式与我沟通。
3. 增加互动，如 “您的看法如何？”
""")
.defaultTools(new CmsAiTools())
.build();
public ChatClient chatClient(ChatClient.Builder builder, CmsAiProperties properties) {
builder.defaultSystem("""
## 人物设定
你是我的知识库AI助手，你把我当作朋友，耐心真诚地回复我提出的相关问题。
你需要遵循以下原则，与关注者进行友善而有价值的沟通。
## 表达方式：
1. 使用简体中文回答我的问题。
2. 使用幽默有趣的方式与我沟通。
3. 增加互动，如 “您的看法如何？”
4. 可以用表情，避免过多表情。
""");
if (properties.getToolCalls()) {
builder.defaultTools(new CmsAiTools());
}
return builder.build();
}
// @Bean

129
modules/cms-ai/src/main/java/com/jeesite/modules/cms/ai/config/WebClientThinkConfig.java Normal file
View File

@@ -0,0 +1,129 @@
/**
* Copyright (c) 2013-Now http://jeesite.com All rights reserved.
* No deletion without permission, or be held responsible to law.
*/
package com.jeesite.modules.cms.ai.config;
import com.jeesite.common.mapper.JsonMapper;
import org.apache.commons.lang3.StringUtils;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.springframework.boot.autoconfigure.condition.ConditionalOnMissingBean;
import org.springframework.boot.web.reactive.function.client.WebClientCustomizer;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;
import org.springframework.core.io.buffer.DataBuffer;
import org.springframework.core.io.buffer.DataBufferUtils;
import org.springframework.core.io.buffer.DefaultDataBufferFactory;
import org.springframework.http.HttpHeaders;
import org.springframework.web.reactive.function.client.ClientResponse;
import org.springframework.web.reactive.function.client.ExchangeFilterFunction;
import reactor.core.publisher.Flux;
import reactor.core.publisher.Mono;
import java.nio.charset.StandardCharsets;
import java.util.ArrayList;
import java.util.List;
import java.util.Map;
import java.util.concurrent.atomic.AtomicBoolean;
/**
* 推理模型OpenAI兼容处理
* @author ThinkGem
*/
@Configuration
public class WebClientThinkConfig {
private final Logger logger = LoggerFactory.getLogger(WebClientThinkConfig.class);
@Bean
@ConditionalOnMissingBean
public WebClientCustomizer webClientCustomizerThink() {
return webClientBuilder -> {
ExchangeFilterFunction requestFilter = ExchangeFilterFunction.ofRequestProcessor(clientRequest -> {
logger.trace("Request url: {}: {}", clientRequest.method(), clientRequest.url());
return Mono.just(clientRequest);
});
ExchangeFilterFunction responseFilter = ExchangeFilterFunction.ofResponseProcessor(clientResponse -> {
logger.trace("Response status: {}", clientResponse.statusCode());
AtomicBoolean thinkingFlag = new AtomicBoolean(false);
Flux<DataBuffer> modifiedBody = clientResponse.bodyToFlux(DataBuffer.class)
.map(buf -> {
byte[] bytes = new byte[buf.readableByteCount()];
buf.read(bytes);
DataBufferUtils.release(buf);
return new String(bytes, StandardCharsets.UTF_8);
})
.flatMap(eventString -> {
logger.trace("Original response: ==> {}", eventString);
List<String> lines = new ArrayList<>();
String[] list = eventString.split("\\n", -1);
for (String line : list) {
if (!line.startsWith("data: ")) {
lines.add(line);
continue;
}
String jsonPart = line.substring("data: ".length()).trim();
if (!(StringUtils.startsWith(jsonPart, "{")
&& StringUtils.endsWith(jsonPart, "}")
&& !"data: [DONE]".equals(line))) {
lines.add(line);
continue;
}
Map<String, Object> map = JsonMapper.fromJson(jsonPart, Map.class);
if (map == null) {
lines.add(line);
continue;
}
// 修改内容字段
List<Object> choices = (List<Object>)map.get("choices");
if (choices == null) {
lines.add(line);
continue;
}
for (Object o : choices) {
Map<String, Object> choice = (Map<String, Object>) o;
if (choice == null) {
continue;
}
Map<String, Object> delta = (Map<String, Object>) choice.get("delta");
if (delta == null) {
continue;
}
String reasoningContent = (String) delta.get("reasoning_content");
String content = (String) delta.get("content");
if (reasoningContent != null) {
if (!thinkingFlag.get()) {
thinkingFlag.set(true);
delta.put("content", "<think>\n" + reasoningContent);
} else {
delta.put("content", reasoningContent);
}
} else {
if (thinkingFlag.get()) {
thinkingFlag.set(false);
delta.put("content", "</think>" + (content == null ? "" : content));
}
}
}
// 重新生成事件字符串
lines.add("data: " + JsonMapper.toJson(map));
}
String finalLine = StringUtils.join(lines, "\n");
logger.trace("Modified response: ==> {}", finalLine);
return Mono.just(finalLine);
})
.map(str -> {
byte[] bytes = str.getBytes(StandardCharsets.UTF_8);
return new DefaultDataBufferFactory().wrap(bytes);
});
ClientResponse modifiedResponse = ClientResponse.from(clientResponse)
.headers(headers -> headers.remove(HttpHeaders.CONTENT_LENGTH))
.body(modifiedBody)
.build();
return Mono.just(modifiedResponse);
});
webClientBuilder.filter(requestFilter).filter(responseFilter);
};
}
}

17
modules/cms-ai/src/main/java/com/jeesite/modules/cms/ai/properties/CmsAiProperties.java Normal file
View File

@@ -0,0 +1,17 @@
package com.jeesite.modules.cms.ai.properties;
import org.springframework.boot.context.properties.ConfigurationProperties;
@ConfigurationProperties("spring.ai")
public class CmsAiProperties {
private Boolean toolCalls = false;
public Boolean getToolCalls() {
return toolCalls;
}
public void setToolCalls(Boolean toolCalls) {
this.toolCalls = toolCalls;
}
}

55
modules/cms-ai/src/main/java/com/jeesite/modules/cms/ai/service/CmsAiChatService.java
View File

@@ -11,18 +11,23 @@ import com.jeesite.common.lang.DateUtils;
import com.jeesite.common.lang.StringUtils;
import com.jeesite.common.service.BaseService;
import com.jeesite.modules.sys.utils.UserUtils;
import jakarta.servlet.http.HttpServletRequest;
import org.springframework.ai.chat.client.ChatClient;
import org.springframework.ai.chat.client.advisor.MessageChatMemoryAdvisor;
import org.springframework.ai.chat.client.advisor.QuestionAnswerAdvisor;
import org.springframework.ai.chat.client.advisor.vectorstore.QuestionAnswerAdvisor;
import org.springframework.ai.chat.memory.ChatMemory;
import org.springframework.ai.chat.messages.AssistantMessage;
import org.springframework.ai.chat.messages.Message;
import org.springframework.ai.chat.messages.UserMessage;
import org.springframework.ai.chat.model.ChatResponse;
import org.springframework.ai.chat.model.Generation;
import org.springframework.ai.vectorstore.SearchRequest;
import org.springframework.ai.vectorstore.VectorStore;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.stereotype.Service;
import org.springframework.web.reactive.function.client.WebClientResponseException;
import reactor.core.publisher.Flux;
import reactor.core.publisher.SignalType;
import java.util.List;
import java.util.Map;
@@ -35,6 +40,8 @@ import java.util.Map;
public class CmsAiChatService extends BaseService {
private static final String CMS_CHAT_CACHE = "cmsChatCache";
private static final String[] USER_MESSAGE_SEARCH = new String[]{"{", "}", "$", "%"};
private static final String[] USER_MESSAGE_REPLACE = new String[]{"\\{", "\\}", "\\$", "\\%"};
@Autowired
private ChatClient chatClient;
@@ -102,14 +109,52 @@ public class CmsAiChatService extends BaseService {
* 聊天对话，流输出
* @author ThinkGem
*/
public Flux<ChatResponse> chatStream(String conversationId, String message) {
public Flux<ChatResponse> chatStream(String conversationId, String message, HttpServletRequest request) {
return chatClient.prompt()
.messages(new UserMessage(message))
.messages(
new UserMessage(StringUtils.replaceEach(message, USER_MESSAGE_SEARCH, USER_MESSAGE_REPLACE))
)
.advisors(
new MessageChatMemoryAdvisor(chatMemory, conversationId, 1024),
new QuestionAnswerAdvisor(vectorStore, SearchRequest.builder().similarityThreshold(0.6F).topK(6).build()))
new QuestionAnswerAdvisor(vectorStore, SearchRequest.builder().similarityThreshold(0.6F).topK(6).build())
)
.stream()
.chatResponse();
.chatResponse()
.doOnNext(response -> {
if (response.getResult() != null && StringUtils.isNotBlank(response.getResult().getOutput().getText())) {
AssistantMessage assistantMessage = (AssistantMessage)request.getAttribute("assistantMessage");
AssistantMessage currAssistantMessage = response.getResult().getOutput();
if (assistantMessage == null) {
request.setAttribute("assistantMessage", currAssistantMessage);
} else {
request.setAttribute("assistantMessage", new AssistantMessage(
assistantMessage.getText() + currAssistantMessage.getText(),
currAssistantMessage.getMetadata()));
}
}
})
.doFinally((signalType) -> {
if (signalType != SignalType.ON_COMPLETE) {
AssistantMessage assistantMessage = (AssistantMessage)request.getAttribute("assistantMessage");
if (assistantMessage != null) {
chatMemory.add(conversationId, assistantMessage);
} else if (signalType == SignalType.CANCEL) {
chatMemory.add(conversationId, new AssistantMessage(text("暂无消息，你已主动停止响应。")));
}
}
})
.onErrorResume(error -> {
String errorMessage = error.getMessage();
if (error instanceof WebClientResponseException webClientError) {
errorMessage = webClientError.getResponseBodyAsString();
}
AssistantMessage assistantMessage = new AssistantMessage(errorMessage);
chatMemory.add(conversationId, assistantMessage);
logger.error("Error message: {}", errorMessage);
return Flux.just(ChatResponse.builder()
.generations(List.of(new Generation(assistantMessage)))
.build());
});
}
}

5
modules/cms-ai/src/main/java/com/jeesite/modules/cms/ai/web/CmsAiChatController.java
View File

@@ -7,6 +7,7 @@ package com.jeesite.modules.cms.ai.web;
import com.jeesite.common.config.Global;
import com.jeesite.common.web.BaseController;
import com.jeesite.modules.cms.ai.service.CmsAiChatService;
import jakarta.servlet.http.HttpServletRequest;
import org.springframework.ai.chat.messages.Message;
import org.springframework.ai.chat.model.ChatResponse;
import org.springframework.beans.factory.annotation.Autowired;
@@ -77,8 +78,8 @@ public class CmsAiChatController extends BaseController {
* @author ThinkGem
*/
@RequestMapping(value = "/stream", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
public Flux<ChatResponse> stream(String id, String message) {
return cmsAiChatService.chatStream(id, message);
public Flux<ChatResponse> stream(String id, String message, HttpServletRequest request) {
return cmsAiChatService.chatStream(id, message, request);
}
}

114
modules/cms-ai/src/main/resources/config/jeesite-cms-ai.yml
View File

@@ -13,19 +13,17 @@ spring:
#api-key: ${BAILIAN_APP_KEY}
# 聊天对话模型
chat:
enabled: false
options:
model: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
#model: DeepSeek-R1-Distill-Qwen-14B
#model: deepseek-r1-distill-llama-8b
max-tokens: 1024
temperature: 0.6
top-p: 0.7
top-p: 0.9
frequency-penalty: 0
logprobs: true
#logprobs: true
# 向量库知识库模型（注意：不同的模型维度不同）
embedding:
enabled: false
options:
model: BAAI/bge-m3
#model: bge-large-zh-v1.5
@@ -33,12 +31,14 @@ spring:
#model: text-embedding-v3
#dimensions: 1024
# 是否启用工具调用
tool-calls: false
# 本地大模型配置（使用该模型，请开启 enabled 参数）
ollama:
base-url: http://localhost:11434
# 聊天对话模型
chat:
enabled: true
options:
model: qwen2.5
#model: deepseek-r1:7b
@@ -48,7 +48,6 @@ spring:
frequency-penalty: 0
# 向量库知识库模型（注意：不同的模型维度不同）
embedding:
enabled: true
# 维度 dimensions 设置为 384
#model: all-minilm:33m
# 维度 dimensions 设置为 768
@@ -68,67 +67,64 @@ spring:
#collection-name: vector_store
collection-name: vector_store_1024
# # Postgresql 向量数据库（PG 连接配置，见下文，需要手动建表）
# pgvector:
# id-type: TEXT
# index-type: HNSW
# distance-type: COSINE_DISTANCE
# initialize-schema: false
# #table-name: vector_store_384
# #dimensions: 384
# #table-name: vector_store_786
# #dimensions: 768
# table-name: vector_store_1024
# dimensions: 1024
# batching-strategy: TOKEN_COUNT
# max-document-batch-size: 10000
# Postgresql 向量数据库（PG 连接配置，见下文，需要手动建表）
pgvector:
id-type: TEXT
index-type: HNSW
distance-type: COSINE_DISTANCE
initialize-schema: false
#table-name: vector_store_384
#dimensions: 384
#table-name: vector_store_786
#dimensions: 768
table-name: vector_store_1024
dimensions: 1024
max-document-batch-size: 10000
# # ES 向量数据库（ES 连接配置，见下文）
# elasticsearch:
# index-name: vector-index
# initialize-schema: true
# dimensions: 1024
# similarity: cosine
# batching-strategy: TOKEN_COUNT
# ES 向量数据库（ES 连接配置，见下文）
elasticsearch:
index-name: vector-index
initialize-schema: true
dimensions: 1024
similarity: cosine
# # Milvus 向量数据库（字符串长度不超过65535）
# milvus:
# client:
# host: "localhost"
# port: 19530
# username: "root"
# password: "milvus"
# initialize-schema: true
# database-name: "default2"
# collection-name: "vector_store2"
# embedding-dimension: 384
# index-type: HNSW
# metric-type: COSINE
# Milvus 向量数据库
milvus:
client:
host: "localhost"
port: 19530
username: "root"
password: "milvus"
initialize-schema: true
database-name: "default"
collection-name: "vector_store"
embedding-dimension: 384
index-type: HNSW
metric-type: COSINE
# ========= Postgresql 向量数据库数据源 =========
#jdbc:
# ds_pgvector:
# type: postgresql
# driver: org.postgresql.Driver
# url: jdbc:postgresql://127.0.0.1:5433/jeesite-ai
# username: postgres
# password: postgres
# testSql: SELECT 1
# pool:
# init: 0
# minIdle: 0
# breakAfterAcquireFailure: true
jdbc:
ds_pgvector:
type: postgresql
driver: org.postgresql.Driver
url: jdbc:postgresql://127.0.0.1:5433/jeesite-ai
username: postgres
password: postgres
testSql: SELECT 1
pool:
init: 0
minIdle: 0
breakAfterAcquireFailure: true
# ========= ES 向量数据库连接配置 =========
#spring.elasticsearch:
# enabled: true
# socket-timeout: 120s
# connection-timeout: 120s
# uris: http://127.0.0.1:9200
# username: elastic
# password: elastic
spring.elasticsearch:
socket-timeout: 120s
connection-timeout: 120s
uris: http://127.0.0.1:9200
username: elastic
password: elastic
# 对话消息存缓存，可自定义存数据库
j2cache: