feat(evaluation): 改进RAG评估功能并优化UI布局

- 添加json-repair依赖以增强JSON解析容错能力
- 移除评估标签中的Beta标识，提升用户体验
- 增加评估问题生成的最大尝试次数
- 优化评估结果页面布局，将检索配置和评估报告并排显示
- 添加任务中心状态同步功能
- 改进指标显示样式和间距

pyproject.toml

@@ -62,6 +62,7 @@ dependencies = [
     "aiofiles>=24.1.0",
     "aiohttp>=3.9.0",
     "deepagents>=0.2.5",
+    "json-repair>=0.54.0",
 ]
 [tool.ruff]
 line-length = 120  # 代码最大行宽

src/services/evaluation_service.py

@@ -287,7 +287,9 @@ class EvaluationService:
         await context.set_progress(25, "生成样本")
 
         with open(data_file_path, "w", encoding="utf-8") as f:
-            while generated < count and attempts < count * 2:
+            # Allow more attempts to generate enough questions
+            max_attempts = max(count * 5, 50)
+            while generated < count and attempts < max_attempts:
                 attempts += 1
                 i0 = random.randrange(len(all_chunks))
                 e0 = embeddings[i0]
@@ -320,20 +322,27 @@ class EvaluationService:
                 try:
                     resp = await asyncio.to_thread(llm.call, prompt, False)
                     content = resp.content if resp else ""
-                    obj = json.loads(content)
+
+                    import json_repair
+                    obj = json_repair.loads(content)
                     q = obj.get("query")
                     a = obj.get("gold_answer")
                     gids = obj.get("gold_chunk_ids")
                     if not q or not a or not isinstance(gids, list):
+                        logger.warning(f"Generated JSON missing fields or invalid format: {obj}")
                         continue
+
                     gids = [str(x) for x in gids if str(x) in allowed_ids]
                     if not gids:
+                        logger.warning("Generated gold_chunk_ids not found in allowed context")
                         continue
+
                     line = {"query": q, "gold_chunk_ids": gids, "gold_answer": a}
                     f.write(json.dumps(line, ensure_ascii=False) + "\n")
                     generated += 1
                     await context.set_progress(0 + int(99 * generated / max(count, 1)), f"已生成 {generated}/{count}")
-                except Exception:
+                except Exception as e:
+                    logger.warning(f"Benchmark generation failed for one item: {e}")
                     continue
 
         meta = {

web/src/components/RAGEvaluationTab.vue

@@ -9,7 +9,7 @@
           <a-select
             v-model:value="selectedBenchmarkId"
             placeholder="请选择评估基准"
-            style="width: 180px"
+            style="width: 240px"
             @change="onBenchmarkChanged"
             :loading="benchmarksLoading"
           >
@@ -202,48 +202,53 @@
         </a-descriptions-item>
       </a-descriptions>
 
+      <!-- 检索配置和整体评估报告 -->
+      <a-row :gutter="16" style="margin-bottom: 20px">
         <!-- 检索配置 -->
-      <a-card size="small" title="检索配置" style="margin-bottom: 20px" v-if="selectedResult.retrieval_config">
+        <a-col :span="12" v-if="selectedResult.retrieval_config">
+          <a-card size="small" title="检索配置">
             <div class="json-viewer-container">
               <pre class="json-viewer">{{ JSON.stringify(selectedResult.retrieval_config, null, 2) }}</pre>
             </div>
           </a-card>
+        </a-col>
 
         <!-- 整体评估报告 -->
-      <div class="evaluation-report">
+        <a-col :span="selectedResult.retrieval_config ? 12 : 24">
-        <h4 style="margin-bottom: 16px">整体评估报告</h4>
+          <a-card size="small" title="整体评估报告">
-        <a-row :gutter="[16, 16]">
+            <!-- 检索指标 -->
-          <a-col :span="12">
+            <div style="margin-bottom: 20px;">
-            <a-card size="small" title="检索指标">
+              <h5 style="margin-bottom: 12px; font-size: 14px; font-weight: 500;">检索指标</h5>
               <div v-if="Object.keys(evaluationStats.retrievalMetrics || {}).length > 0">
                 <div v-for="(value, key) in evaluationStats.retrievalMetrics" :key="key" class="report-metric">
-                  <span class="metric-label">{{ getMetricTitle(key) }}</span>
+                  <span class="metric-label">{{ getMetricTitle(key) }}：</span>
                   <span class="metric-value" :style="{ color: getScoreColor(value) }">
                     {{ formatMetricValue(value) }}
                   </span>
                 </div>
               </div>
               <span v-else class="no-metrics">-</span>
-            </a-card>
+            </div>
-          </a-col>
+
-          <a-col :span="12">
+            <!-- 答案准确性 -->
-            <a-card size="small" title="答案准确性">
+            <div>
+              <h5 style="margin-bottom: 12px; font-size: 14px; font-weight: 500;">答案准确性</h5>
               <div class="accuracy-stats">
                 <div class="accuracy-item">
-                  <span class="accuracy-label">正确答案数</span>
+                  <span class="accuracy-label">正确答案数：</span>
                   <span class="accuracy-value">{{ evaluationStats.correctAnswers || 0 }} / {{ evaluationStats.totalQuestions || 0 }}</span>
                 </div>
                 <div class="accuracy-item">
-                  <span class="accuracy-label">准确率</span>
+                  <span class="accuracy-label">准确率：</span>
                   <span class="accuracy-value" :style="{ color: getScoreColor(evaluationStats.answerAccuracy) }">
                     {{ (evaluationStats.answerAccuracy * 100).toFixed(1) }}%
                   </span>
                 </div>
               </div>
+            </div>
           </a-card>
         </a-col>
       </a-row>
-      </div>
 
       <!-- 详细结果表格 -->
       <h4 style="margin-bottom: 16px">详细评估结果</h4>
@@ -317,6 +322,7 @@ import { evaluationApi } from '@/apis/knowledge_api';
 import ModelSelectorComponent from '@/components/ModelSelectorComponent.vue';
 import SearchConfigModal from './SearchConfigModal.vue';
 import { SettingOutlined, ReloadOutlined } from '@ant-design/icons-vue';
+import { useTaskerStore } from '@/stores/tasker';
 
 const props = defineProps({
   databaseId: {
@@ -327,6 +333,9 @@ const props = defineProps({
 
 const emit = defineEmits(['switch-to-benchmarks']);
 
+// 使用任务中心 store
+const taskerStore = useTaskerStore();
+
 // 状态
 const selectedBenchmarkId = ref(null);
 const selectedBenchmark = ref(null);
@@ -539,6 +548,8 @@ const startEvaluation = async () => {
     if (response.message === 'success') {
       message.success('评估任务已开始');
       loadEvaluationHistory();
+      // 刷新任务中心的任务列表
+      taskerStore.loadTasks();
     } else {
       message.error(response.message || '启动评估失败');
     }
@@ -1178,6 +1189,7 @@ onMounted(() => {
 
     .metric-label {
       font-size: 14px;
+      padding-right: 18px;
       color: var(--gray-700);
     }
 

web/src/views/DataBaseInfoView.vue

@@ -42,14 +42,14 @@
             ref="mindmapSectionRef"
           />
         </a-tab-pane>
-        <a-tab-pane key="evaluation" tab="RAG评估（Beta）">
+        <a-tab-pane key="evaluation" tab="RAG评估">
           <RAGEvaluationTab
             v-if="databaseId"
             :database-id="databaseId"
             @switch-to-benchmarks="activeTab = 'benchmarks'"
           />
         </a-tab-pane>
-        <a-tab-pane key="benchmarks" tab="评估基准管理（Beta）">
+        <a-tab-pane key="benchmarks" tab="评估基准">
           <div class="benchmark-management-container">
             <div class="benchmark-content">
               <EvaluationBenchmarks