1.13.3

deploy/erupt-cloud-server-docker/Dockerfile

@@ -1,5 +1,5 @@
 # 默认使用阿里云镜像
-FROM openjdk:11-jre-slim
+FROM eclipse-temurin:17-jre
 
 # 设置工作目录
 WORKDIR /app

deploy/erupt-cloud-server-docker/pom.xml

@@ -6,7 +6,7 @@
     <parent>
         <groupId>xyz.erupt</groupId>
         <artifactId>erupt</artifactId>
-        <version>1.13.2</version>
+        <version>1.13.3</version>
         <relativePath>../../pom.xml</relativePath>
     </parent>
 

erupt-admin/pom.xml

@@ -10,7 +10,7 @@
     <parent>
         <groupId>xyz.erupt</groupId>
         <artifactId>erupt</artifactId>
-        <version>1.13.2</version>
+        <version>1.13.3</version>
         <relativePath>../pom.xml</relativePath>
     </parent>
 

erupt-ai-web/package.json

@@ -23,6 +23,7 @@
     "markdown-it-footnote": "^4.0.0",
     "markdown-it-highlightjs": "^4.2.0",
     "markdown-it-ins": "^4.0.0",
+    "markdown-it-katex": "^2.0.3",
     "markdown-it-link-attributes": "^4.0.1",
     "markdown-it-mark": "^4.0.0",
     "markdown-it-mermaid": "^0.2.5",

erupt-ai-web/src/App.vue

@@ -127,7 +127,6 @@ const send = (message: string) => {
     eventSource.onmessage = (event) => {
       sending.value = false;
       sendDisabled.value = true;
-      console.log(JSON.parse(event.data).text)
       accumulatedMarkdown.value += JSON.parse(event.data).text;
       let msg = messages.value[messages.value.length - 1];
       if (msg.loading) {

erupt-ai-web/src/app.less

@@ -148,4 +148,9 @@
     background: #ddd;
     margin: @base-margin * 2 0;
   }
+
+  /* 修复 KaTeX 上下标显示问题 */
+  .katex .vlist {
+    vertical-align: baseline !important;
+  }
 }

erupt-ai-web/src/components/markdown.ts

@@ -11,34 +11,37 @@ import abbr from 'markdown-it-abbr';
 import deflist from 'markdown-it-deflist';
 import linkAttributes from 'markdown-it-link-attributes';
 import taskLists from 'markdown-it-task-lists';
-import katex from 'katex';
+import markdownItKatex from 'markdown-it-katex';
 import 'katex/dist/katex.min.css';
+import mermaid from 'mermaid';
 
+// 初始化 mermaid
+// if (typeof window !== 'undefined') {
+//     mermaid.initialize({
+//         startOnLoad: true,
+//         theme: 'default',
+//         securityLevel: 'loose'
+//     });
+// }
 
-// 自定义数学公式渲染
-// @ts-ignore
-const renderMath = (md) => {
-    const defaultRender = md.renderer.rules.text;
-    md.renderer.rules.text = (tokens: {
-        [x: string]: any;
-    }, idx: string | number, options: any, env: any, self: any) => {
-        const token = tokens[idx];
-        const match = token.content.match(/(\$\$?)([\s\S]*?)\1/g);
-        if (match) {
-            match.forEach((m: string) => {
-                const isBlock = m.startsWith('$$');
-                const tex = m.slice(2, -2);
-                try {
-                    const html = katex.renderToString(tex, {displayMode: isBlock});
-                    token.content = token.content.replace(m, html);
-                } catch (e: any) {
-                    token.content = token.content.replace(m, `<span class="error">Error rendering math: ${e.message}</span>`);
-                }
-            });
-        }
-        return defaultRender(tokens, idx, options, env, self);
-    };
-};
+function preprocessLatex(text: string): string {
+    // 1. 先将 ```latex 代码块的内容提取出来（保留原始内容）
+    text = text.replace(/```latex\s*([\s\S]*?)```/g, (_match, content) => {
+        return '\n' + content.trim() + '\n';
+    });
+
+    // 2. 统一将 \[ ... \] 转换为 $$...$$（LaTeX 块级公式）
+    text = text.replace(/\\\[([\s\S]*?)\\\]/g, (_match, formula) => {
+        return '$$' + formula.trim() + '$$';
+    });
+
+    // 3. 将 \( ... \) 转换为 $...$（LaTeX 行内公式）
+    text = text.replace(/\\\(([\s\S]*?)\\\)/g, (_match, formula) => {
+        return '$' + formula.trim() + '$';
+    });
+
+    return text;
+}
 
 // 创建 markdown-it 实例并加载插件
 const md = new MarkdownIt({
@@ -47,6 +50,12 @@ const md = new MarkdownIt({
     linkify: true,
     typographer: true,
     highlight: (str: string, lang: string) => {
+        // 处理 mermaid 图表
+        if (lang === 'mermaid') {
+            return `<div class="mermaid">${str}</div>`;
+        }
+
+        // 处理代码高亮
         if (lang && hljs.getLanguage(lang)) {
             return `<pre class="hljs"><code>${hljs.highlight(str, {language: lang}).value}</code></pre>`;
         }
@@ -63,6 +72,26 @@ const md = new MarkdownIt({
     .use(deflist)
     .use(linkAttributes, {attrs: {target: '_blank', rel: 'noopener'}})
     .use(taskLists)
-    .use(renderMath);
+    .use(markdownItKatex, {
+        throwOnError: false,
+        errorColor: '#cc0000'
+    });
+
+// 重写 render 方法，添加预处理
+const originalRender = md.render.bind(md);
+md.render = function (text: string, env?: any): string {
+    const preprocessedText = preprocessLatex(text);
+    const html = originalRender(preprocessedText, env);
+
+    if (html.indexOf('<div class="mermaid">') !== -1) {
+        setTimeout(() => {
+            mermaid.run({
+                querySelector: '.mermaid'
+            });
+        }, 0);
+    }
+
+    return html;
+};
 
 export default md;

erupt-ai/pom.xml

@@ -10,7 +10,7 @@
     <parent>
         <groupId>xyz.erupt</groupId>
         <artifactId>erupt</artifactId>
-        <version>1.13.2</version>
+        <version>1.13.3</version>
         <relativePath>../pom.xml</relativePath>
     </parent>
 

erupt-ai/src/main/java/xyz/erupt/ai/config/AiProp.java

@@ -19,4 +19,6 @@ public class AiProp {
 
     private boolean devMode = false;
 
+    private boolean enableFunctionCall = true;
+
 }

erupt-ai/src/main/java/xyz/erupt/ai/core/OpenAi.java

@@ -20,6 +20,7 @@
 import java.io.IOException;
 import java.util.HashMap;
 import java.util.List;
+import java.util.concurrent.TimeUnit;
 import java.util.function.Consumer;
 
 /**
@@ -30,6 +31,14 @@
 @Slf4j
 public abstract class OpenAi extends LlmCore {
 
+    private final OkHttpClient client = new OkHttpClient().newBuilder()
+            .connectTimeout(30, TimeUnit.SECONDS)
+            .readTimeout(5, TimeUnit.MINUTES)    // 流式需要很长！
+            .writeTimeout(30, TimeUnit.SECONDS)
+            .pingInterval(10, TimeUnit.SECONDS)  // 保持连接心跳
+            .build();
+    ;
+
     private final Gson gson = new GsonBuilder().create();
 
     public String chatApiPath() {
@@ -50,7 +59,6 @@ public LlmConfig config() {
     public ChatCompletionResponse chat(LlmRequest llmRequest, String userPrompt, List<ChatCompletionMessage> assistantPrompt) {
         assistantPrompt.add(new ChatCompletionMessage(MessageRole.user, userPrompt));
         ChatCompletion completion = ChatCompletion.builder().model(llmRequest.getModel()).stream(false).messages(assistantPrompt).build();
-        OkHttpClient client = new OkHttpClient();
         RequestBody body = RequestBody.create(
                 gson.toJson(completion),
                 MediaType.parse("application/json; charset=utf-8")
@@ -77,6 +85,7 @@ public ChatCompletionResponse chat(LlmRequest llmRequest, String userPrompt, Lis
     @Override
     @SneakyThrows
     public void chatSse(LlmRequest llmRequest, String userPrompt, List<ChatCompletionMessage> assistantPrompt, Consumer<SseListener> listener) {
+        assistantPrompt.removeIf(it -> StringUtils.isBlank(it.getContent()));
         ChatCompletion completion = ChatCompletion.builder().model(llmRequest.getModel()).messages(assistantPrompt).stream(true).build();
         completion.setResponse_format(new HashMap<>() {{
             this.put("type", String.valueOf(llmRequest.getResponseFormat()));
@@ -118,8 +127,8 @@ public void onResponse(@NotNull Call call, @NotNull Response response) throws IO
                             String line = source.readUtf8Line();
                             if (StringUtils.isNotBlank(line)) {
                                 if (!response.isSuccessful()) {
-                                    this.onFailure(call, new IOException(response.body().string()));
-                                    log.error("Failed to get llm response from server: {}", response.body());
+                                    this.onFailure(call, new IOException(response.body().string() + line));
+                                    log.error("Failed to get llm response from server: {}", response.body() + " → " + line);
                                     return;
                                 } else {
                                     try {
@@ -146,7 +155,8 @@ public void onResponse(@NotNull Call call, @NotNull Response response) throws IO
                                             listener.accept(sseListener);
                                         }
                                     } catch (Exception e) {
-                                        this.onFailure(call, new IOException(e));
+                                        this.onFailure(call, new IOException(e + "→" + line));
+                                        break;
                                     }
                                 }
                             }

erupt-ai/src/main/java/xyz/erupt/ai/llm/OpenRouter.java

@@ -1,7 +1,6 @@
 package xyz.erupt.ai.llm;
 
 import org.springframework.stereotype.Component;
-import xyz.erupt.ai.core.LlmConfig;
 import xyz.erupt.ai.core.OpenAi;
 
 @Component
@@ -17,4 +16,8 @@ public String api() {
         return "https://openrouter.ai";
     }
 
+    @Override
+    public String chatApiPath() {
+        return "/api/v1/chat/completions";
+    }
 }

erupt-ai/src/main/java/xyz/erupt/ai/llm/QWen.java → erupt-ai/src/main/java/xyz/erupt/ai/llm/Qwen.java

@@ -1,15 +1,14 @@
 package xyz.erupt.ai.llm;
 
 import org.springframework.stereotype.Component;
-import xyz.erupt.ai.core.LlmConfig;
 import xyz.erupt.ai.core.OpenAi;
 
 /**
  * @author YuePeng
  * date 2025/2/26 22:58
  */
 @Component
-public class QWen extends OpenAi {
+public class Qwen extends OpenAi {
 
     @Override
     public String chatApiPath() {

erupt-ai/src/main/java/xyz/erupt/ai/model/LLM.java

@@ -59,7 +59,8 @@ public class LLM extends MetaModelUpdateVo {
                     type = EditType.CHOICE,
                     notNull = true,
                     search = @Search,
-                    choiceType = @ChoiceType(fetchHandler = LlmCore.H.class, trigger = LLMDataProxy.class)
+                    onchange = LLMDataProxy.class,
+                    choiceType = @ChoiceType(fetchHandler = LlmCore.H.class)
             )
     )
     private String llm;

erupt-ai/src/main/java/xyz/erupt/ai/model/LLMDataProxy.java

@@ -7,14 +7,13 @@
 import org.springframework.transaction.annotation.Transactional;
 import xyz.erupt.ai.config.AiProp;
 import xyz.erupt.ai.core.LlmCore;
-import xyz.erupt.annotation.fun.ChoiceTrigger;
 import xyz.erupt.annotation.fun.DataProxy;
 import xyz.erupt.annotation.fun.OperationHandler;
 import xyz.erupt.annotation.sub_erupt.Tpl;
+import xyz.erupt.annotation.sub_field.sub_edit.OnChange;
 import xyz.erupt.jpa.dao.EruptDao;
 import xyz.erupt.linq.lambda.LambdaSee;
 
-import java.util.Collections;
 import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
@@ -24,7 +23,7 @@
  * date 2025/3/1 18:19
  */
 @Component
-public class LLMDataProxy implements DataProxy<LLM>, Tpl.TplHandler, ChoiceTrigger, OperationHandler<LLM, Void> {
+public class LLMDataProxy implements DataProxy<LLM>, Tpl.TplHandler, OnChange<LLM>, OperationHandler<LLM, Void> {
 
     @Resource
     private EruptDao eruptDao;
@@ -44,19 +43,6 @@ public void bindTplData(Map<String, Object> binding, String[] params) {
 
     public static Gson gson = new GsonBuilder().setPrettyPrinting().serializeNulls().create();
 
-    @Override
-    public Map<String, Object> trigger(Object code, String[] params) {
-        if (null != code && !"null".equals(code)) {
-            Map<String, Object> ret = new HashMap<>();
-            ret.put(LambdaSee.field(LLM::getModel), LlmCore.getLLM(code.toString()).model());
-            ret.put(LambdaSee.field(LLM::getApiUrl), LlmCore.getLLM(code.toString()).api());
-            ret.put(LambdaSee.field(LLM::getApiKey), "");
-            ret.put(LambdaSee.field(LLM::getConfig), gson.toJson(LlmCore.getLLM(code.toString()).config()));
-            return ret;
-        }
-        return Collections.emptyMap();
-    }
-
     @Override
     @Transactional
     public String exec(List<LLM> data, Void o, String[] param) {
@@ -67,4 +53,21 @@ public String exec(List<LLM> data, Void o, String[] param) {
         eruptDao.find(LLM.class, data.get(0).getId()).setDefaultLLM(true);
         return "";
     }
+
+    @Override
+    public Map<String, Object> populateForm(LLM llm, String[] params) {
+        if (null != llm.getLlm()) {
+            Map<String, Object> ret = new HashMap<>();
+            ret.put(LambdaSee.field(LLM::getModel), LlmCore.getLLM(llm.getLlm()).model());
+            ret.put(LambdaSee.field(LLM::getApiUrl), LlmCore.getLLM(llm.getLlm()).api());
+            ret.put(LambdaSee.field(LLM::getConfig), gson.toJson(LlmCore.getLLM(llm.getLlm()).config()));
+            return ret;
+        }
+        return Map.of();
+    }
+
+    @Override
+    public Map<String, String> buildEditExpr(LLM llm, String[] params) {
+        return Map.of();
+    }
 }

erupt-ai/src/main/java/xyz/erupt/ai/service/LLMService.java

@@ -3,6 +3,7 @@
 import jakarta.annotation.Resource;
 import lombok.SneakyThrows;
 import lombok.extern.slf4j.Slf4j;
+import org.apache.commons.lang3.StringUtils;
 import org.springframework.http.MediaType;
 import org.springframework.scheduling.annotation.Async;
 import org.springframework.stereotype.Service;
@@ -83,7 +84,9 @@ public List<ChatCompletionMessage> geneCompletionPrompt(Chat chat, LLMAgent llmA
             }
         } else {
             // Function Call
-            chatCompletionMessages.add(new ChatCompletionMessage(MessageRole.system, aiFunctionManager.getFunctionCallPrompt()));
+            if (aiProp.isEnableFunctionCall()) {
+                chatCompletionMessages.add(new ChatCompletionMessage(MessageRole.system, aiFunctionManager.getFunctionCallPrompt()));
+            }
         }
         List<ChatMessage> chatMessages = eruptDao.lambdaQuery(ChatMessage.class)
                 .eq(ChatMessage::getChatId, chat.getId())
@@ -110,9 +113,18 @@ public void sendSse(MetaContext metaContext, LLMAgent llmAgent, SseEmitter emitt
             llm.chatSse(llmRequest, chatMessage.getContent(), completionMessage, it -> {
                 if (it.isFinish()) {
                     String msg = it.getOutput().toString();
-                    if (it.getOutput().toString().length() <= MESSAGE_TS || it.isError()) {
-                        msg = this.sendMessage(emitter, it.getOutput().toString(), llmModal, chatMessage, completionMessage);
+                    // finish 时，无论之前发送了多少，都要确保最后的增量被发送
+                    if (msg.length() > MESSAGE_TS && !it.isError()) {
+                        // 只发送最后的增量部分
+                        if (!StringUtils.isEmpty(it.getCurrMessage())) {
+                            this.sendMessage(emitter, it.getCurrMessage(), llmModal, chatMessage, completionMessage);
+                        }
+                    } else {
+                        // 短消息直接发送完整内容
+                        msg = this.sendMessage(emitter, msg, llmModal, chatMessage, completionMessage);
                     }
+
+                    // 保存到数据库时使用完整消息
                     chatMessage.setTokens((long) it.getUsage().getPrompt_tokens());
                     eruptDao.mergeAndFlush(chatMessage);
                     eruptDao.persistAndFlush(ChatMessage.create(chatMessage.getChatId(), llmModal.getLlm(), llmModal.getModel(), ChatSenderType.MODEL, msg, (long) it.getUsage().getCompletion_tokens()));
@@ -138,7 +150,7 @@ public void sendSse(MetaContext metaContext, LLMAgent llmAgent, SseEmitter emitt
     }
 
     private String sendMessage(SseEmitter emitter, String userMessage, LLM llm, ChatMessage chatMessage, List<ChatCompletionMessage> userContext) {
-        if (aiFunctionManager.exist(userMessage.trim())) {
+        if (aiProp.isEnableFunctionCall() && aiFunctionManager.exist(userMessage.trim())) {
             String functionMessage = aiFunctionManager.call(userMessage, llm, chatMessage.getContent(), userContext);
             try {
                 emitter.send(GsonFactory.getGson().toJson(new SseBody(functionMessage)), MediaType.TEXT_EVENT_STREAM);