feat: 评测模块新增详情页与创建向导

新增 EvalTaskDetail 与逐样本结果类型，Mock 与适配器补充评测详情接口，路由注册详情页并将维度创建并入向导；EvalCreateView 改为分步向导（任务配置、规则设置、维度表单），新增 EvalDetailView 展示综合得分与样本判定，列表补充详情入口，配套两份回归脚本。

frontend/package.json

@@ -14,6 +14,8 @@
     "test:data-process-wizard": "node scripts/regression-data-process-wizard.mjs",
     "test:dataset-task-tab": "node scripts/regression-dataset-task-tab.mjs",
     "test:dataset-preview": "node scripts/regression-dataset-preview.mjs",
+    "test:eval-create": "node scripts/regression-eval-create-wizard.mjs",
+    "test:eval-detail": "node scripts/regression-eval-detail.mjs",
     "test:model-manage": "node scripts/regression-model-manage.mjs",
     "test:training-log-layout": "node scripts/regression-training-log-layout.mjs",
     "test:page-surface": "node scripts/regression-page-surface.mjs"

frontend/scripts/regression-eval-create-wizard.mjs

@@ -0,0 +1,148 @@
+import assert from 'node:assert/strict'
+import { readFile } from 'node:fs/promises'
+import { fileURLToPath } from 'node:url'
+import path from 'node:path'
+
+const scriptDir = path.dirname(fileURLToPath(import.meta.url))
+const sourceRoot = path.resolve(scriptDir, '../src')
+const evalViewDir = path.join(sourceRoot, 'views/eval')
+
+const [
+  createSource,
+  taskStepSource,
+  ruleStepSource,
+  listSource,
+  routerSource,
+  apiSource,
+] = await Promise.all([
+  readFile(path.join(evalViewDir, 'EvalCreateView.vue'), 'utf8'),
+  readFile(path.join(evalViewDir, 'create/EvalTaskSetupStep.vue'), 'utf8'),
+  readFile(path.join(evalViewDir, 'create/EvalRuleSetupStep.vue'), 'utf8'),
+  readFile(path.join(evalViewDir, 'EvalView.vue'), 'utf8'),
+  readFile(path.join(sourceRoot, 'router/index.ts'), 'utf8'),
+  readFile(path.join(sourceRoot, 'api/modules/eval.ts'), 'utf8'),
+])
+
+function extractRouteBlock(source, routePath) {
+  const pathPattern = new RegExp(`path:\\s*['"]${routePath.replaceAll('/', '\\/')}['"]`)
+  const pathIndex = source.search(pathPattern)
+  assert.notEqual(pathIndex, -1, `未找到路由 ${routePath}`)
+
+  const openBrace = source.lastIndexOf('{', pathIndex)
+  assert.notEqual(openBrace, -1, `路由 ${routePath} 缺少左花括号`)
+
+  let depth = 0
+  for (let index = openBrace; index < source.length; index += 1) {
+    if (source[index] === '{') depth += 1
+    if (source[index] === '}') depth -= 1
+    if (depth === 0) return source.slice(openBrace + 1, index)
+  }
+
+  assert.fail(`路由 ${routePath} 缺少右花括号`)
+}
+
+const createDimensionRoute = extractRouteBlock(routerSource, 'model-eval/dimension/create')
+assert.match(
+  createDimensionRoute,
+  /redirect:\s*['"]\/model-eval\/create['"]/,
+  '独立创建维度入口必须收敛到新建评测向导',
+)
+assert.doesNotMatch(
+  createDimensionRoute,
+  /DimensionCreateView/,
+  '独立创建维度路由不应继续加载旧创建页',
+)
+
+const editDimensionRoute = extractRouteBlock(routerSource, 'model-eval/dimension/:id/edit')
+assert.match(
+  editDimensionRoute,
+  /DimensionCreateView\.vue/,
+  '编辑维度路由必须继续复用 DimensionCreateView.vue',
+)
+
+const dimensionTabStart = listSource.indexOf('<!-- 评测维度 -->')
+assert.notEqual(dimensionTabStart, -1, '评测列表缺少维度标签页')
+const dimensionTabSource = listSource.slice(dimensionTabStart)
+assert.doesNotMatch(dimensionTabSource, /create-text=["']添加维度["']/, '维度标签页不应保留“添加维度”按钮')
+assert.doesNotMatch(dimensionTabSource, /@create=["']handleCreateClick["']/, '维度标签页不应保留独立创建事件')
+
+assert.match(
+  apiSource,
+  /createDimension\s*=\s*\([^)]*\)\s*=>\s*[\s\S]*?post<\{\s*id:\s*string\s*\|\s*number\s*\}>\(['"]\/dimension['"]\s*,\s*data\)/,
+  'createDimension 必须声明返回包含 string | number 类型 id',
+)
+
+assert.match(createSource, /EvalTaskSetupStep/, '向导缺少任务配置子组件')
+assert.match(createSource, /EvalRuleSetupStep/, '向导缺少评测规则子组件')
+assert.match(
+  createSource,
+  /任务配置[\s\S]*?评测规则/,
+  '评测创建向导必须按“任务配置、评测规则”定义两个步骤',
+)
+assert.match(createSource, /currentStep\s*=\s*ref\(0\)/, '评测创建向导缺少当前步骤状态')
+assert.match(createSource, /class=["']custom-wizard-steps["']/, '评测向导未复用数据处理的自定义步骤条结构')
+assert.match(createSource, /class=["']step-connector["']/, '评测向导步骤条缺少连接线')
+assert.match(createSource, /:aria-current=/, '步骤条缺少当前步骤的可访问性绑定')
+assert.match(createSource, /currentStep === index \? 'step' : undefined/, '步骤条当前步骤语义不正确')
+assert.doesNotMatch(createSource, /<el-steps\b|<el-step\b/, '评测向导不应继续使用 Element Plus 默认步骤条')
+assert.match(
+  createSource,
+  /<EvalTaskSetupStep[\s\S]*?v-if=["'][^"']*currentStep[^"']*0[^"']*["']/,
+  '任务配置子组件没有绑定第一步',
+)
+assert.match(
+  createSource,
+  /<EvalRuleSetupStep[\s\S]*?v-else/,
+  '评测规则子组件没有绑定第二步',
+)
+
+assert.match(taskStepSource, /prop=["']dataset_id["']/, '数据集字段缺少表单校验标识')
+assert.match(
+  `${createSource}\n${taskStepSource}`,
+  /data_source[\s\S]*?dataset[\s\S]*?dataset_id/,
+  '选择评测数据集时必须校验 dataset_id',
+)
+
+assert.match(createSource, /['"]baseline['"]/, '向导缺少基线评测分支')
+assert.match(
+  `${createSource}\n${ruleStepSource}`,
+  /已有维度|existing/,
+  '评测规则步骤缺少使用已有维度的分支',
+)
+assert.match(
+  `${createSource}\n${ruleStepSource}`,
+  /新建(?:评测)?(?:维度|规则)|create/,
+  '评测规则步骤缺少新建维度的分支',
+)
+
+assert.match(
+  createSource,
+  /await\s+createDimension\([\s\S]*?await\s+startEval\(/,
+  '最终提交必须先创建新维度，再使用维度 id 启动评测',
+)
+assert.match(
+  createSource,
+  /createdDimension(?:\.value)?\.id|createdDimensionId|dimensionResult\.id/,
+  '启动评测前必须读取新建维度返回的 id',
+)
+assert.match(
+  createSource,
+  /eval_type[\s\S]*?baseline[\s\S]*?(?:dimension_id|createDimension)/,
+  '基线评测必须有明确的不创建自定义维度分支',
+)
+assert.match(
+  createSource,
+  /ruleMode[\s\S]*?['"]existing['"]/,
+  '已有维度模式必须有独立的提交分支',
+)
+
+assert.match(createSource, /:loading=["']submitting["']/, '最终提交按钮缺少 loading 状态')
+assert.match(createSource, /:disabled=["'][^"']*submitting/, '提交期间必须禁用重复操作')
+assert.match(createSource, /class=["'][^"']*wizard-footer/, '评测向导缺少统一底部操作栏')
+assert.match(
+  createSource,
+  /\.wizard-footer\s*\{[\s\S]*?display:\s*flex[\s\S]*?justify-content:\s*flex-end/,
+  '底部操作栏必须保持企业表单常用的右对齐布局',
+)
+
+console.log('评测创建向导回归检查通过')

frontend/scripts/regression-eval-detail.mjs

@@ -0,0 +1,99 @@
+import assert from 'node:assert/strict'
+import { readFile } from 'node:fs/promises'
+import { fileURLToPath } from 'node:url'
+import path from 'node:path'
+
+const scriptDir = path.dirname(fileURLToPath(import.meta.url))
+const sourceRoot = path.resolve(scriptDir, '../src')
+
+const [routerSource, listSource, detailSource, apiSource, mockSource] = await Promise.all([
+  readFile(path.join(sourceRoot, 'router/index.ts'), 'utf8'),
+  readFile(path.join(sourceRoot, 'views/eval/EvalView.vue'), 'utf8'),
+  readFile(path.join(sourceRoot, 'views/eval/EvalDetailView.vue'), 'utf8'),
+  readFile(path.join(sourceRoot, 'api/modules/eval.ts'), 'utf8'),
+  readFile(path.join(sourceRoot, 'mock/adapter.ts'), 'utf8'),
+])
+
+function extractRouteBlock(source, routePath) {
+  const pathPattern = new RegExp(`path:\\s*['"]${routePath.replaceAll('/', '\\/')}['"]`)
+  const pathIndex = source.search(pathPattern)
+  assert.notEqual(pathIndex, -1, `未找到路由 ${routePath}`)
+
+  const openBrace = source.lastIndexOf('{', pathIndex)
+  assert.notEqual(openBrace, -1, `路由 ${routePath} 缺少左花括号`)
+
+  let depth = 0
+  for (let index = openBrace; index < source.length; index += 1) {
+    if (source[index] === '{') depth += 1
+    if (source[index] === '}') depth -= 1
+    if (depth === 0) return source.slice(openBrace + 1, index)
+  }
+
+  assert.fail(`路由 ${routePath} 缺少右花括号`)
+}
+
+const detailRoute = extractRouteBlock(routerSource, 'model-eval/:id')
+assert.match(detailRoute, /name:\s*['"]model-eval-detail['"]/, '评测详情路由名称不正确')
+assert.match(
+  detailRoute,
+  /component:\s*\(\)\s*=>\s*import\(['"]@\/views\/eval\/EvalDetailView\.vue['"]\)/,
+  '评测详情路由未加载 EvalDetailView.vue',
+)
+
+assert.match(
+  listSource,
+  /router\.push\(\s*\{\s*name:\s*['"]model-eval-detail['"]\s*,\s*params:\s*\{\s*id:\s*row\.id\s*\}\s*\}\s*\)/,
+  '评测任务详情按钮必须使用命名路由并携带任务 id',
+)
+assert.doesNotMatch(listSource, /详情功能开发中/, '评测任务详情入口仍是占位提示')
+
+assert.match(apiSource, /export\s+const\s+getEvalDetail\b/, '评测 API 缺少 getEvalDetail')
+assert.match(
+  apiSource,
+  /get(?:<[^>]+>)?\(`\/model-eval\/\$\{id\}`\)/,
+  'getEvalDetail 未请求 GET /model-eval/:id',
+)
+assert.ok(
+  mockSource.includes('/^\\/model-eval\\/([^/]+)$/'),
+  'Mock 未按任务 id 匹配 /model-eval/:id',
+)
+assert.match(mockSource, /method\s*===\s*['"]get['"]/, '评测详情 Mock 未处理 GET 请求')
+
+for (const className of [
+  'overall-review',
+  'score-hero',
+  'improvement-list',
+  'sample-results-table',
+  'dimension-summary',
+]) {
+  assert.match(
+    detailSource,
+    new RegExp(`class=["'][^"']*\\b${className}\\b`),
+    `评测详情页缺少 .${className} 结构`,
+  )
+}
+
+for (const label of ['输入', '标准答案', '模型回答', '得分', '判定']) {
+  assert.match(
+    detailSource,
+    new RegExp(`(?:label=["'][^"']*${label}[^"']*["']|>${label}<)`),
+    `样本结果列表缺少“${label}”语义`,
+  )
+}
+
+assert.match(detailSource, /\bloading\b/, '评测详情页缺少加载状态')
+assert.match(detailSource, /\berror\b/, '评测详情页缺少错误状态')
+assert.match(
+  detailSource,
+  /(?:el-empty|empty-text|样本结果为空|暂无样本|暂无数据)/,
+  '评测详情页缺少空状态',
+)
+assert.match(detailSource, /<el-pagination\b/, '评测详情页缺少样本分页')
+
+for (const marketingLabel of ['OVERALL SCORE', 'LLM REVIEW', 'DIMENSIONS', 'SAMPLE RESULTS']) {
+  assert.doesNotMatch(detailSource, new RegExp(marketingLabel), `企业详情页不应保留展示型英文标签：${marketingLabel}`)
+}
+assert.doesNotMatch(detailSource, /linear-gradient\s*\(/, '企业详情页不应使用大面积渐变背景')
+assert.match(detailSource, /grid-template-columns:\s*repeat\(4,\s*minmax\(0,\s*1fr\)\)/, '桌面端概览或维度区应采用紧凑四列布局')
+
+console.log('评测详情页回归检查通过')

frontend/src/api/modules/eval.ts

@@ -1,9 +1,12 @@
 import { get, post, put, del } from '../request'
-import type { EvalTask, Dimension } from '@/types'
+import type { EvalTask, EvalTaskDetail, Dimension } from '@/types'
 
 /** 评测任务列表 */
 export const getEvalList = () => get<EvalTask[]>('/model-eval')
 
+/** 评测任务详情 */
+export const getEvalDetail = (id: string | number) => get<EvalTaskDetail>(`/model-eval/${id}`)
+
 /** 删除评测任务 */
 export const deleteEval = (id: string | number) => del(`/model-eval/${id}`)
 
@@ -17,7 +20,8 @@ export const getDimensionList = () => get<Dimension[]>('/dimension')
 export const getDimension = (id: string | number) => get<Dimension>(`/dimension/${id}`)
 
 /** 创建维度 */
-export const createDimension = (data: Partial<Dimension>) => post('/dimension', data)
+export const createDimension = (data: Partial<Dimension>) =>
+  post<{ id: string | number }>('/dimension', data)
 
 /** 编辑维度 */
 export const updateDimension = (id: string | number, data: Partial<Dimension>) =>

frontend/src/mock/adapter.ts

@@ -16,6 +16,7 @@ import {
   mockFineTuneList,
   mockCompareList,
   mockEvalList,
+  mockEvalDetails,
   mockDimensions,
   mockLogFiles,
   mockTrainingLogFiles,
@@ -229,6 +230,12 @@ async function handleMock(config: AxiosRequestConfig) {
   if (url === '/model-eval' && method === 'get') return ok(mockEvalList)
   if (url === '/model-eval' && method === 'delete') return ok({ deleted: true })
   if (url === '/model-eval/start' && method === 'post') return ok({ task_id: Math.floor(Math.random() * 1000) + 1 })
+  m = url.match(/^\/model-eval\/([^/]+)$/)
+  const evalTaskId = m?.[1]
+  if (evalTaskId && method === 'get') {
+    const found = mockEvalDetails.find((item) => String(item.id) === evalTaskId)
+    return found ? ok(found) : fail('评测任务不存在', 404)
+  }
   if (url === '/dimension' && method === 'get') return ok(mockDimensions)
   if (url === '/dimension' && method === 'post') {
     return ok({ id: Math.floor(Math.random() * 1000) + 100 })

frontend/src/mock/data.ts

@@ -10,6 +10,7 @@ import type {
   DatasetItem,
   CompareTask,
   EvalTask,
+  EvalTaskDetail,
   Dimension,
   SystemInfo,
   HealthMetrics,
@@ -254,6 +255,139 @@ export const mockEvalList: EvalTask[] = [
   { id: 3, eval_task_name: '基线对比评测', model_name: 'Qwen2.5-7B-Instruct', dataset: '金融评测集', metric: 'accuracy', score: 72.3, status: 'running', create_time: '2026-02-10T09:00:00Z' },
 ]
 
+export const mockEvalDetails: EvalTaskDetail[] = [
+  {
+    ...mockEvalList[0],
+    evaluator_model: 'GPT-4o',
+    sample_count: 3,
+    completed_count: 3,
+    passed_count: 3,
+    completed_time: '2026-02-01T10:18:00Z',
+    overall_score: 87.5,
+    overall_score_max: 100,
+    overall_evaluation: '模型在金融基础知识与常见计算题上表现稳定，答案整体准确、直接。',
+    improvement_suggestions: ['补充期限匹配、再平衡等资产配置知识', '在建议类回答中明确假设与风险边界', '增加复杂金融场景和反事实样本训练'],
+    dimension_summary: [
+      { name: '核心事实正确性', score: 91, max_score: 100, pass_rate: 100 },
+      { name: '信息完整性', score: 84, max_score: 100, pass_rate: 100 },
+      { name: '无幻觉', score: 90, max_score: 100, pass_rate: 100 },
+      { name: '格式合规性', score: 85, max_score: 100, pass_rate: 100 },
+    ],
+    samples: [
+      {
+        id: 'finance-1', index: 1, status: 'completed', score: 1, max_score: 1, passed: true, judgement: '正确',
+        input: '某公司流动资产为 800 万元，流动负债为 500 万元，流动比率是多少？',
+        reference_answer: '流动比率为 1.6。',
+        model_output: '流动比率 = 流动资产 ÷ 流动负债 = 800 ÷ 500 = 1.6。',
+        evaluation_reason: '计算过程与结论均正确，并给出了必要公式。',
+        dimension_scores: [{ name: '准确性', score: 1, max_score: 1 }, { name: '清晰度', score: 1, max_score: 1 }],
+      },
+      {
+        id: 'finance-2', index: 2, status: 'completed', score: 1, max_score: 1, passed: true, judgement: '正确',
+        input: '简述央行提高存款准备金率对市场流动性的影响。',
+        reference_answer: '商业银行可贷资金减少，市场流动性通常收紧。',
+        model_output: '提高存款准备金率会减少银行可用于放贷的资金，通常使市场流动性收紧。',
+        evaluation_reason: '准确覆盖了传导路径，表述简洁。',
+        dimension_scores: [{ name: '准确性', score: 1, max_score: 1 }, { name: '完整性', score: 0.9, max_score: 1 }],
+      },
+      {
+        id: 'finance-3', index: 3, status: 'completed', score: 0.8, max_score: 1, passed: true, judgement: '部分正确',
+        input: '投资者风险承受能力较低时，资产配置应注意什么？',
+        reference_answer: '降低高波动资产比例，重视分散配置、流动性与本金安全。',
+        model_output: '应以低波动资产为主，分散投资并预留流动资金，同时控制权益类资产比例。',
+        evaluation_reason: '建议方向正确，但可进一步说明期限匹配与再平衡机制。',
+        dimension_scores: [{ name: '准确性', score: 0.9, max_score: 1 }, { name: '完整性', score: 0.7, max_score: 1 }],
+      },
+    ],
+  },
+  {
+    ...mockEvalList[1],
+    evaluator_model: 'Claude-3.5-Sonnet',
+    sample_count: 3,
+    completed_count: 3,
+    passed_count: 2,
+    completed_time: '2026-02-05T11:21:00Z',
+    overall_score: 0.82,
+    overall_score_max: 1,
+    overall_evaluation: '模型具备较好的客服语气和基础问题处理能力，但对订单状态约束的识别仍不稳定。',
+    improvement_suggestions: ['增加已发货、已出库等边界状态样本', '要求模型在承诺操作前先核验业务规则', '为每类客服场景补充标准处理步骤'],
+    dimension_summary: [
+      { name: '核心事实正确性', score: 0.86, max_score: 1, pass_rate: 67 },
+      { name: '信息完整性', score: 0.78, max_score: 1, pass_rate: 67 },
+      { name: '无幻觉', score: 0.7, max_score: 1, pass_rate: 67 },
+      { name: '格式合规性', score: 0.94, max_score: 1, pass_rate: 100 },
+    ],
+    samples: [
+      {
+        id: 'service-1', index: 1, status: 'completed', score: 0.91, max_score: 1, passed: true, judgement: '正确',
+        input: '用户反馈收到的商品外包装破损，应如何回复？',
+        reference_answer: '表达歉意，核实订单及破损情况，并提供补发或退款方案。',
+        model_output: '很抱歉给您带来不便。请提供订单号和破损照片，我们核实后可为您安排补发或退款。',
+        evaluation_reason: '关键信息覆盖完整，语气符合客服场景。',
+        dimension_scores: [{ name: '相关性', score: 0.95, max_score: 1 }, { name: '服务语气', score: 0.96, max_score: 1 }],
+      },
+      {
+        id: 'service-2', index: 2, status: 'completed', score: 0.84, max_score: 1, passed: true, judgement: '部分正确',
+        input: '用户询问订单为何仍未发货。',
+        reference_answer: '致歉并核查订单状态，告知预计发货时间和后续查询方式。',
+        model_output: '抱歉让您久等了，我可以帮您核查订单状态，并尽快反馈预计发货时间。',
+        evaluation_reason: '态度友好，但未主动说明后续查询方式。',
+        dimension_scores: [{ name: '服务语气', score: 0.95, max_score: 1 }, { name: '完整性', score: 0.72, max_score: 1 }],
+      },
+      {
+        id: 'service-3', index: 3, status: 'completed', score: 0.69, max_score: 1, passed: false, judgement: '错误', error_type: '其他',
+        input: '用户要求立即取消已出库的订单。',
+        reference_answer: '说明已出库订单可能无法直接取消，协助拦截或引导拒收、退货。',
+        model_output: '好的，我现在为您取消订单，请稍候。',
+        evaluation_reason: '未说明已出库限制，承诺了可能无法完成的操作。',
+        dimension_scores: [{ name: '相关性', score: 0.7, max_score: 1 }, { name: '规则遵循', score: 0.35, max_score: 1 }],
+      },
+    ],
+  },
+  {
+    ...mockEvalList[2],
+    evaluator_model: 'GPT-4o',
+    sample_count: 3,
+    completed_count: 2,
+    passed_count: 1,
+    overall_score: 72.3,
+    overall_score_max: 100,
+    overall_evaluation: '当前为阶段性评价：模型基础计算能力正常，但逻辑推理表现波动，任务仍在运行。',
+    improvement_suggestions: ['补充形式逻辑与多步推理样本', '任务完成后再依据完整结果复核综合结论'],
+    dimension_summary: [
+      { name: '核心事实正确性', score: 66.7, max_score: 100, pass_rate: 50 },
+      { name: '信息完整性', score: 58, max_score: 100, pass_rate: 50 },
+      { name: '无幻觉', score: 88, max_score: 100, pass_rate: 100 },
+      { name: '格式合规性', score: 76, max_score: 100, pass_rate: 50 },
+    ],
+    samples: [
+      {
+        id: 'baseline-1', index: 1, status: 'completed', score: 1, max_score: 1, passed: true, judgement: '正确',
+        input: '计算 15% 的 240。',
+        reference_answer: '36。',
+        model_output: '240 × 15% = 36。',
+        evaluation_reason: '计算与结果均正确。',
+        dimension_scores: [{ name: '准确性', score: 1, max_score: 1 }, { name: '清晰度', score: 1, max_score: 1 }],
+      },
+      {
+        id: 'baseline-2', index: 2, status: 'completed', score: 0, max_score: 1, passed: false, judgement: '错误', error_type: '其他',
+        input: '若所有 A 都是 B，且某个 C 是 A，可以推出什么？',
+        reference_answer: '该 C 是 B。',
+        model_output: '无法确定 C 是否属于 B。',
+        evaluation_reason: '未正确应用包含关系的传递性。',
+        dimension_scores: [{ name: '准确性', score: 0, max_score: 1 }, { name: '逻辑性', score: 0, max_score: 1 }],
+      },
+      {
+        id: 'baseline-3', index: 3, status: 'pending', score: null, max_score: 1,
+        input: '用一句话解释什么是机会成本。',
+        reference_answer: '机会成本是选择某方案时放弃的最佳替代方案的价值。',
+        model_output: '',
+        evaluation_reason: '等待模型生成与评分。',
+      },
+    ],
+  },
+]
+
 export const mockDimensions: Dimension[] = [
   { id: 1, name: '回答准确性', type: 'classification', description: '评估模型回答是否准确', eval_model: 'GPT-4o', eval_method: 'standard', eval_prompt: '# 角色\n你是专业的评估专家...', is_active: true, is_default: true, create_time: '2025-12-01T08:00:00Z' },
   { id: 2, name: '综合评分', type: 'metric', description: '0-5 分综合评分', eval_model: 'Claude-3.5-Sonnet', eval_method: 'metric_standard', eval_prompt: '# 角色\n你是专业评分专家...', is_active: true, is_default: false, score_min: 0, score_max: 5, pass_threshold: 3.5, create_time: '2025-12-05T09:00:00Z' },

frontend/src/router/index.ts

@@ -46,11 +46,16 @@ const routes: RouteRecordRaw[] = [
         component: () => import('@/views/eval/EvalCreateView.vue'),
         meta: { title: '新建评测' },
       },
+      {
+        path: 'model-eval/:id',
+        name: 'model-eval-detail',
+        component: () => import('@/views/eval/EvalDetailView.vue'),
+        meta: { title: '评测详情' },
+      },
       {
         path: 'model-eval/dimension/create',
         name: 'dimension-create',
-        component: () => import('@/views/eval/DimensionCreateView.vue'),
+        redirect: '/model-eval/create',
-        meta: { title: '添加评测维度' },
       },
       {
         path: 'model-eval/dimension/:id/edit',

frontend/src/types/index.ts

@@ -155,6 +155,7 @@ export interface CompareModelRef {
 // ============ 模型评测 ============
 export type EvalType = 'custom' | 'baseline'
 export type DimensionType = 'classification' | 'metric' | 'text_similarity'
+export type EvalErrorType = '混淆' | '不完整' | '格式偏差' | '幻觉' | '其他'
 
 export interface EvalTask {
   id: number | string
@@ -168,6 +169,47 @@ export interface EvalTask {
   create_time?: string
 }
 
+/** 单个测试样本的评测结果 */
+export interface EvalSampleResult {
+  id: number | string
+  index: number
+  input: string
+  reference_answer?: string
+  model_output: string
+  score: number | null
+  max_score?: number
+  passed?: boolean
+  status: 'completed' | 'pending' | 'failed'
+  judgement?: '正确' | '部分正确' | '错误'
+  evaluation_reason?: string
+  error_type?: EvalErrorType
+  dimension_scores?: Array<{
+    name: string
+    score: number
+    max_score: number
+  }>
+}
+
+/** 评测任务详情，包含逐样本结果和综合评价 */
+export interface EvalTaskDetail extends EvalTask {
+  evaluator_model?: string
+  sample_count: number
+  completed_count: number
+  passed_count?: number
+  completed_time?: string
+  overall_score: number
+  overall_score_max: number
+  overall_evaluation: string
+  improvement_suggestions: string[]
+  dimension_summary: Array<{
+    name: string
+    score: number
+    max_score: number
+    pass_rate: number
+  }>
+  samples: EvalSampleResult[]
+}
+
 export interface Dimension {
   id: number | string
   name: string

frontend/src/views/eval/DimensionCreateView.vue

@@ -1,19 +1,17 @@
 <script setup lang="ts">
-import { ref, reactive, computed, watch, onMounted } from 'vue'
+import { ref, reactive, computed, onMounted } from 'vue'
 import { useRoute, useRouter } from 'vue-router'
 import { ElMessage, type FormInstance, type FormRules } from 'element-plus'
-import { MdEditor } from 'md-editor-v3'
-import 'md-editor-v3/lib/style.css'
 import PageCard from '@/components/PageCard.vue'
+import DimensionFormFields, { type DimensionFormDraft } from './create/DimensionFormFields.vue'
 import {
   createDimension,
   updateDimension,
   getDimension,
 } from '@/api/modules/eval'
 import { getModelList } from '@/api/modules/model'
-import { EVAL_METHODS, EVAL_METHOD_PROMPTS } from '@/constants/dimension'
+import { EVAL_METHODS } from '@/constants/dimension'
-import { DIMENSION_TYPE_MAP } from '@/constants'
+import type { ModelItem } from '@/types'
-import type { DimensionType, ModelItem } from '@/types'
 
 const route = useRoute()
 const router = useRouter()
@@ -24,9 +22,9 @@ const editId = computed(() => route.params.id as string | undefined)
 
 const evalModels = ref<ModelItem[]>([])
 
-const form = reactive({
+const form = reactive<DimensionFormDraft>({
   name: '',
-  type: '' as DimensionType | '',
+  type: '',
   description: '',
   eval_model: '',
   eval_method: '' as string | string[],
@@ -49,44 +47,10 @@ const rules: FormRules = {
   ],
   type: [{ required: true, message: '请选择指标类型', trigger: 'change' }],
   eval_model: [{ required: true, message: '请选择大模型', trigger: 'change' }],
+  eval_method: [{ required: true, message: '请选择评估方式', trigger: 'change' }],
   eval_prompt: [{ required: true, message: '请填写评估 Prompt', trigger: 'blur' }],
 }
 
-const typeOptions = Object.entries(DIMENSION_TYPE_MAP).map(([value, label]) => ({ value, label }))
-
-/** 当前类型下的评估方法选项 */
-const currentMethods = computed(() => (form.type ? EVAL_METHODS[form.type] || [] : []))
-
-/** 是否文本相似度类型（多选评估方法） */
-const isTextSimilarity = computed(() => form.type === 'text_similarity')
-
-/** 切换指标类型时重置相关字段 */
-watch(
-  () => form.type,
-  (t) => {
-    if (!t) return
-    const methods = EVAL_METHODS[t] || []
-    const first = methods[0]?.value || ''
-    form.eval_method = isTextSimilarity.value ? [first] : first
-    // 切换 Prompt
-    if (form.type !== 'text_similarity' && EVAL_METHOD_PROMPTS[first]) {
-      form.eval_prompt = EVAL_METHOD_PROMPTS[first]
-    }
-  },
-)
-
-/** 评估方法变化时更新 Prompt */
-watch(
-  () => form.eval_method,
-  (m) => {
-    if (form.type === 'text_similarity' || !m) return
-    const method = Array.isArray(m) ? m[0] : m
-    if (method && EVAL_METHOD_PROMPTS[method]) {
-      form.eval_prompt = EVAL_METHOD_PROMPTS[method]
-    }
-  },
-)
-
 async function loadEditData() {
   if (!editId.value) return
   try {
@@ -176,100 +140,7 @@ onMounted(() => {
 <template>
   <PageCard :title="isEdit ? '编辑评测维度' : '添加评测维度'">
     <el-form ref="formRef" :model="form" :rules="rules" label-width="130px" style="max-width: 760px">
-      <el-form-item label="维度名称" prop="name">
+      <DimensionFormFields v-model="form" :eval-models="evalModels" />
-        <el-input v-model="form.name" placeholder="请输入维度名称" maxlength="50" show-word-limit />
-      </el-form-item>
-
-      <el-form-item label="指标类型" prop="type">
-        <el-select v-model="form.type" placeholder="请选择指标类型" style="width: 100%">
-          <el-option v-for="opt in typeOptions" :key="opt.value" :label="opt.label" :value="opt.value" />
-        </el-select>
-      </el-form-item>
-
-      <el-form-item label="描述">
-        <el-input v-model="form.description" type="textarea" :rows="2" maxlength="200" show-word-limit />
-      </el-form-item>
-
-      <!-- 分类型 / 指标型：需要选择大模型和评估方法 -->
-      <template v-if="form.type === 'classification' || form.type === 'metric'">
-        <el-form-item label="选择大模型" prop="eval_model">
-          <el-select v-model="form.eval_model" placeholder="请选择评测模型" filterable style="width: 100%">
-            <el-option v-for="m in evalModels" :key="m.id" :label="m.name" :value="m.name" />
-          </el-select>
-        </el-form-item>
-
-        <el-form-item label="评估方式" prop="eval_method">
-          <el-radio-group v-model="(form.eval_method as string)">
-            <el-radio v-for="m in currentMethods" :key="m.value" :value="m.value">
-              {{ m.name }}
-              <span class="method-desc">{{ m.desc }}</span>
-            </el-radio>
-          </el-radio-group>
-        </el-form-item>
-
-        <el-form-item label="评估 Prompt" prop="eval_prompt">
-          <MdEditor v-model="form.eval_prompt" language="zh-CN" :toolbars-exclude="['github', 'save']" style="height: 320px" />
-        </el-form-item>
-      </template>
-
-      <!-- 文本相似度：多选评估方法 + BLEU/精度配置 -->
-      <template v-if="form.type === 'text_similarity'">
-        <el-form-item label="评估方式">
-          <el-checkbox-group v-model="(form.eval_method as string[])">
-            <el-checkbox v-for="m in currentMethods" :key="m.value" :value="m.value">
-              {{ m.name }}
-              <span class="method-desc">{{ m.desc }}</span>
-            </el-checkbox>
-          </el-checkbox-group>
-        </el-form-item>
-
-        <el-form-item label="BLEU n-gram">
-          <el-select v-model="form.bleu_n" style="width: 160px">
-            <el-option :value="1" label="1-gram" />
-            <el-option :value="2" label="2-gram" />
-            <el-option :value="3" label="3-gram" />
-            <el-option :value="4" label="4-gram" />
-          </el-select>
-        </el-form-item>
-
-        <el-form-item label="输出精度">
-          <el-select v-model="form.output_precision" style="width: 160px">
-            <el-option :value="1" label="1 位小数" />
-            <el-option :value="2" label="2 位小数" />
-            <el-option :value="3" label="3 位小数" />
-            <el-option :value="4" label="4 位小数" />
-          </el-select>
-        </el-form-item>
-      </template>
-
-      <!-- 指标型：评分范围 + 通过阈值 -->
-      <template v-if="form.type === 'metric'">
-        <el-form-item label="评分最小值">
-          <el-input-number v-model="form.score_min" :min="0" :step="1" style="width: 160px" />
-        </el-form-item>
-        <el-form-item label="评分最大值">
-          <el-input-number v-model="form.score_max" :min="0" :step="1" style="width: 160px" />
-        </el-form-item>
-        <el-form-item label="通过阈值">
-          <div style="width: 100%">
-            <el-slider
-              v-model="form.pass_threshold"
-              :min="form.score_min"
-              :max="form.score_max"
-              :step="0.5"
-              show-input
-              :show-input-controls="false"
-            />
-          </div>
-        </el-form-item>
-      </template>
-
-      <el-form-item label="启用该维度">
-        <el-switch v-model="form.is_active" />
-      </el-form-item>
-      <el-form-item label="设为默认">
-        <el-switch v-model="form.is_default" />
-      </el-form-item>
 
       <el-form-item>
         <el-button type="primary" :loading="submitting" @click="handleSubmit">
@@ -280,11 +151,3 @@ onMounted(() => {
     </el-form>
   </PageCard>
 </template>
-
-<style scoped>
-.method-desc {
-  color: #909399;
-  font-size: 12px;
-  margin-left: 6px;
-}
-</style>

frontend/src/views/eval/EvalCreateView.vue

@@ -1,91 +1,165 @@
 <script setup lang="ts">
-import { ref, reactive, computed, onMounted, watch } from 'vue'
+import { onMounted, ref, watch } from 'vue'
 import { useRouter } from 'vue-router'
-import { ElMessage, type FormInstance, type FormRules } from 'element-plus'
+import { ElMessage } from 'element-plus'
 import PageCard from '@/components/PageCard.vue'
-import { startEval } from '@/api/modules/eval'
+import EvalTaskSetupStep, { type EvalTaskSetupDraft } from './create/EvalTaskSetupStep.vue'
-import { getDimensionList } from '@/api/modules/eval'
+import EvalRuleSetupStep, { type EvalRuleSetupDraft } from './create/EvalRuleSetupStep.vue'
-import { getTrainedModels } from '@/api/modules/model'
+import { createDimension, getDimensionList, startEval } from '@/api/modules/eval'
+import { getTrainedModels, getModelList } from '@/api/modules/model'
 import { getDatasetList } from '@/api/modules/dataset'
 import { getSystemInfo } from '@/api/modules/system'
-import type { Dimension, TrainedModel, DatasetItem, GpuInfo } from '@/types'
+import type { DatasetItem, Dimension, GpuInfo, ModelItem, TrainedModel } from '@/types'
+
+type StepExposed = { validate: () => Promise<boolean> }
 
 const router = useRouter()
-const formRef = ref<FormInstance>()
+const loading = ref(false)
 const submitting = ref(false)
+const taskStepRef = ref<StepExposed>()
+const ruleStepRef = ref<StepExposed>()
 
 const trainedModels = ref<TrainedModel[]>([])
 const evalDatasets = ref<DatasetItem[]>([])
 const dimensions = ref<Dimension[]>([])
+const evalModels = ref<ModelItem[]>([])
 const gpus = ref<GpuInfo[]>([])
+const createdDimensionId = ref<string | number>('')
 
-const form = reactive({
+const taskForm = ref<EvalTaskSetupDraft>({
   eval_task_name: '',
-  eval_type: 'custom' as 'custom' | 'baseline',
+  model_id: '',
-  model_id: '' as string | number,
   gpu_id: '',
-  data_source: 'dataset' as 'dataset' | 'inference',
+  data_source: 'dataset',
-  dataset_id: '' as string | number,
+  dataset_id: '',
-  dimension_id: '' as string | number,
+  leaderboard: false,
-  leaderboard: false, // 修复原项目字段名 bug（原读 leader 恒 false）
 })
 
-const rules: FormRules = {
+const ruleForm = ref<EvalRuleSetupDraft>({
-  eval_task_name: [{ required: true, message: '请输入任务名称', trigger: 'blur' }],
+  ruleMode: 'existing',
-  model_id: [{ required: true, message: '请选择评测模型', trigger: 'change' }],
+  dimension_id: '',
-  gpu_id: [{ required: true, message: '请选择 GPU', trigger: 'change' }],
+  newDimension: {
-  dimension_id: [{ required: true, message: '请选择评测维度', trigger: 'change' }],
+    name: '',
-}
+    type: '',
+    description: '',
+    eval_model: '',
+    eval_method: '',
+    eval_prompt: '',
+    is_active: true,
+    is_default: false,
+    bleu_n: 1,
+    output_precision: 3,
+    score_min: 0,
+    score_max: 5,
+    pass_threshold: 3,
+  },
+})
 
-const showEvalRules = computed(() => form.eval_type === 'custom')
+watch(
-const showDataset = computed(() => form.data_source === 'dataset')
+  () => ruleForm.value.newDimension,
-
+  () => {
-/** 选中维度详情 */
+    createdDimensionId.value = ''
-const selectedDimension = computed(() =>
+  },
-  dimensions.value.find((d) => d.id == form.dimension_id),
+  { deep: true },
 )
 
 async function loadData() {
-  try {
+  loading.value = true
-    const [models, dims, allDs, sys] = await Promise.all([
+  const results = await Promise.allSettled([
     getTrainedModels(),
     getDimensionList(),
     getDatasetList(),
     getSystemInfo(),
+    getModelList(),
   ])
-    trainedModels.value = models?.models || []
+
-    dimensions.value = dims || []
+  if (results[0].status === 'fulfilled') trainedModels.value = results[0].value?.models || []
-    evalDatasets.value = (allDs || []).filter((d: DatasetItem) => d.type === 'eval')
+  if (results[1].status === 'fulfilled') dimensions.value = results[1].value || []
-    gpus.value = sys?.gpu || []
+  if (results[2].status === 'fulfilled') {
-  } catch {
+    evalDatasets.value = (results[2].value || []).filter((dataset) => dataset.type === 'eval')
-    // ignore
   }
+  if (results[3].status === 'fulfilled') gpus.value = results[3].value?.gpu || []
+  if (results[4].status === 'fulfilled') {
+    evalModels.value = (results[4].value || []).filter((model) => model.purpose === 'evaluation')
+  }
+
+  const failedCount = results.filter((result) => result.status === 'rejected').length
+  if (failedCount > 0) {
+    ElMessage.error(`部分基础数据加载失败（${failedCount} 项），请刷新页面后重试`)
+  }
+  loading.value = false
+}
+
+
+
+function buildDimensionPayload() {
+  const draft = ruleForm.value.newDimension
+  const payload: Partial<Dimension> = {
+    name: draft.name,
+    type: draft.type,
+    description: draft.description,
+    eval_model: draft.type === 'text_similarity' ? undefined : draft.eval_model,
+    eval_method: draft.eval_method,
+    eval_prompt: draft.type === 'text_similarity' ? undefined : draft.eval_prompt,
+    is_active: draft.is_active,
+    is_default: draft.is_default,
+    create_time: new Date().toISOString(),
+  }
+  if (draft.type === 'text_similarity') {
+    payload.bleu_n = draft.bleu_n
+    payload.output_precision = draft.output_precision
+  }
+  if (draft.type === 'metric') {
+    payload.score_min = draft.score_min
+    payload.score_max = draft.score_max
+    payload.pass_threshold = draft.pass_threshold
+  }
+  return payload
+}
+
+async function resolveDimensionId() {
+  if (ruleForm.value.ruleMode === 'baseline') return ''
+  if (ruleForm.value.ruleMode === 'existing') return ruleForm.value.dimension_id
+  if (createdDimensionId.value !== '') return createdDimensionId.value
+
+  const created = await createDimension(buildDimensionPayload())
+  if (created?.id === undefined || created.id === null || created.id === '') {
+    throw new Error('评测维度已提交，但未返回维度 ID，无法启动评测任务')
+  }
+  createdDimensionId.value = created.id
+  return created.id
 }
 
 async function handleSubmit() {
-  if (!formRef.value) return
+  if (loading.value || submitting.value) return
-  await formRef.value.validate(async (valid) => {
+  const taskValid = await taskStepRef.value?.validate()
-    if (!valid) return
+  const ruleValid = await ruleStepRef.value?.validate()
+  if (!taskValid || !ruleValid) {
+    ElMessage.warning('请检查并完善所有必填项')
+    return
+  }
+
   submitting.value = true
   try {
+    const dimensionId = await resolveDimensionId()
     await startEval({
-        eval_task_name: form.eval_task_name,
+      eval_task_name: taskForm.value.eval_task_name,
-        eval_type: form.eval_type,
+      eval_type: ruleForm.value.ruleMode === 'baseline' ? 'baseline' : 'custom',
-        model_id: form.model_id,
+      model_id: taskForm.value.model_id,
-        gpu_id: form.gpu_id,
+      gpu_id: taskForm.value.gpu_id,
-        dataset_id: form.dataset_id,
+      dataset_id: taskForm.value.data_source === 'dataset' ? taskForm.value.dataset_id : '',
-        dimension_id: form.dimension_id,
+      dimension_id: dimensionId,
-        data_source: form.data_source,
+      data_source: taskForm.value.data_source,
-        leaderboard: form.leaderboard,
+      leaderboard: taskForm.value.leaderboard,
     })
-      ElMessage.success('评测任务已启动')
+    ElMessage.success('评测任务已创建并启动')
     router.push('/model-eval')
-    } catch {
+  } catch (error) {
-      // ignore
+    const message = error instanceof Error ? error.message : '创建评测任务失败，请稍后重试'
+    ElMessage.error(message)
   } finally {
     submitting.value = false
   }
-  })
 }
 
 function handleCancel() {
@@ -96,82 +170,77 @@ onMounted(loadData)
 </script>
 
 <template>
-  <PageCard title="新建评测">
+  <PageCard title="新建评测任务" subtitle="完成任务配置与评测规则设置后启动评测">
-    <el-form ref="formRef" :model="form" :rules="rules" label-width="120px" style="max-width: 720px">
+    <div v-loading="loading" class="form-container">
-      <el-form-item label="任务名称" prop="eval_task_name">
+      <div class="section-card">
-        <el-input v-model="form.eval_task_name" placeholder="请输入任务名称" maxlength="50" show-word-limit />
+        <div class="section-title">任务基本配置</div>
-      </el-form-item>
+        <EvalTaskSetupStep
-
+          ref="taskStepRef"
-      <el-form-item label="评测方式">
+          v-model="taskForm"
-        <el-radio-group v-model="form.eval_type">
+          :trained-models="trainedModels"
-          <el-radio-button value="custom">自定义评测</el-radio-button>
+          :eval-datasets="evalDatasets"
-          <el-radio-button value="baseline">基线评测</el-radio-button>
+          :gpus="gpus"
-        </el-radio-group>
+          :loading="loading"
-      </el-form-item>
+          :disabled="submitting"
-
-      <el-form-item label="评测模型" prop="model_id">
-        <el-select v-model="form.model_id" placeholder="请选择已训练模型" filterable style="width: 100%">
-          <el-option
-            v-for="m in trainedModels"
-            :key="m.id"
-            :label="m.name"
-            :value="m.id"
-            :disabled="!m.merged"
         />
-        </el-select>
+      </div>
-      </el-form-item>
 
-      <el-form-item label="选择 GPU" prop="gpu_id">
+      <div class="section-card">
-        <el-select v-model="form.gpu_id" placeholder="请选择 GPU" style="width: 100%">
+        <div class="section-title">评测规则设置</div>
-          <el-option v-for="(g, idx) in gpus" :key="idx" :label="`${g.name} (GPU ${idx})`" :value="idx" />
+        <EvalRuleSetupStep
-        </el-select>
+          ref="ruleStepRef"
-      </el-form-item>
+          v-model="ruleForm"
+          :dimensions="dimensions"
+          :eval-models="evalModels"
+          :loading="loading"
+          :disabled="submitting"
+        />
+      </div>
+    </div>
 
-      <el-form-item label="数据来源">
+    <div class="form-footer">
-        <el-radio-group v-model="form.data_source">
+      <el-button :disabled="submitting" @click="handleCancel">取消</el-button>
-          <el-radio value="dataset">评测数据集</el-radio>
+      <el-button
-          <el-radio value="inference">推理结果集</el-radio>
+        type="primary"
-        </el-radio-group>
+        :loading="submitting"
-      </el-form-item>
+        :disabled="loading"
-
+        @click="handleSubmit"
-      <el-form-item v-if="showDataset" label="评测数据集" prop="dataset_id">
-        <el-select v-model="form.dataset_id" placeholder="请选择评测数据集" filterable style="width: 100%">
-          <el-option v-for="d in evalDatasets" :key="d.id" :label="d.name" :value="d.id" />
-        </el-select>
-      </el-form-item>
-
-      <!-- 评测规则（仅 custom 显示） -->
-      <template v-if="showEvalRules">
-        <el-divider content-position="left">评测规则</el-divider>
-        <el-form-item label="评测维度" prop="dimension_id">
-          <el-select
-            v-model="form.dimension_id"
-            placeholder="请选择评测维度"
-            filterable
-            style="width: 100%"
       >
-            <el-option v-for="d in dimensions" :key="d.id" :label="d.name" :value="d.id" />
+        创建并启动评测
-          </el-select>
+      </el-button>
-        </el-form-item>
+    </div>
-
-        <el-form-item v-if="selectedDimension" label="维度详情">
-          <el-descriptions :column="1" border size="small">
-            <el-descriptions-item label="评估模型">{{ selectedDimension.eval_model || '-' }}</el-descriptions-item>
-            <el-descriptions-item label="评估方法">
-              {{ Array.isArray(selectedDimension.eval_method) ? selectedDimension.eval_method.join(', ') : selectedDimension.eval_method || '-' }}
-            </el-descriptions-item>
-          </el-descriptions>
-        </el-form-item>
-      </template>
-
-      <el-form-item label="加入排行榜">
-        <el-switch v-model="form.leaderboard" />
-      </el-form-item>
-
-      <el-form-item>
-        <el-button type="primary" :loading="submitting" @click="handleSubmit">启动评测</el-button>
-        <el-button @click="handleCancel">取消</el-button>
-      </el-form-item>
-    </el-form>
   </PageCard>
 </template>
+
+<style scoped>
+.form-container {
+  display: flex;
+  flex-direction: column;
+  gap: 24px;
+}
+
+.section-card {
+  padding: 24px;
+  border-radius: 8px;
+  background: #f9fafb;
+  border: 1px solid #e5e7eb;
+}
+
+.section-title {
+  font-size: 16px;
+  font-weight: 600;
+  color: #111827;
+  margin-bottom: 24px;
+  padding-bottom: 12px;
+  border-bottom: 1px solid #e5e7eb;
+}
+
+.form-footer {
+  display: flex;
+  justify-content: flex-end;
+  gap: 12px;
+  margin-top: 32px;
+  padding-top: 24px;
+  border-top: 1px solid #e5e7eb;
+}
+</style>

frontend/src/views/eval/EvalDetailView.vue

@@ -0,0 +1,749 @@
+<script setup lang="ts">
+import { computed, onMounted, ref } from 'vue'
+import { useRoute } from 'vue-router'
+import PageCard from '@/components/PageCard.vue'
+import ModelStatusTag from '@/components/ModelStatusTag.vue'
+import { getEvalDetail } from '@/api/modules/eval'
+import type { EvalSampleResult, EvalTaskDetail } from '@/types'
+
+const route = useRoute()
+const taskId = route.params.id as string
+
+const loading = ref(true)
+const loadError = ref('')
+const detail = ref<EvalTaskDetail | null>(null)
+const keyword = ref('')
+const judgementFilter = ref('')
+const currentPage = ref(1)
+const pageSize = ref(10)
+
+const filteredSamples = computed(() => {
+  const normalizedKeyword = keyword.value.trim().toLowerCase()
+  return (detail.value?.samples || []).filter((sample) => {
+    const matchesJudgement = !judgementFilter.value || sample.judgement === judgementFilter.value
+    const searchableText = [
+      sample.input,
+      sample.reference_answer,
+      sample.model_output,
+      sample.evaluation_reason,
+      sample.error_type,
+    ].filter(Boolean).join(' ').toLowerCase()
+    return matchesJudgement && (!normalizedKeyword || searchableText.includes(normalizedKeyword))
+  })
+})
+
+const paginatedSamples = computed(() => {
+  const start = (currentPage.value - 1) * pageSize.value
+  return filteredSamples.value.slice(start, start + pageSize.value)
+})
+
+const completionRate = computed(() => {
+  const total = detail.value?.sample_count || 0
+  return total ? Math.round(((detail.value?.completed_count || 0) / total) * 100) : 0
+})
+
+const passRate = computed(() => {
+  const completed = detail.value?.completed_count || 0
+  return completed ? Math.round(((detail.value?.passed_count || 0) / completed) * 100) : 0
+})
+
+const overallScore = computed(() => formatScore(detail.value?.overall_score, detail.value?.overall_score_max))
+
+function formatDateTime(value?: string) {
+  if (!value) return '-'
+  const date = new Date(value)
+  return Number.isNaN(date.getTime())
+    ? value
+    : date.toLocaleString('zh-CN', { hour12: false })
+}
+
+function formatScore(score?: number | null, maxScore?: number) {
+  if (score == null) return '-'
+  const decimals = Number.isInteger(score) ? 0 : 2
+  return `${score.toFixed(decimals)}${maxScore != null ? ` / ${maxScore}` : ''}`
+}
+
+function judgementTagType(judgement?: EvalSampleResult['judgement']) {
+  if (judgement === '正确') return 'success'
+  if (judgement === '部分正确') return 'warning'
+  if (judgement === '错误') return 'danger'
+  return 'info'
+}
+
+function resetPage() {
+  currentPage.value = 1
+}
+
+async function loadDetail() {
+  loading.value = true
+  loadError.value = ''
+  try {
+    detail.value = await getEvalDetail(taskId)
+  } catch {
+    detail.value = null
+    loadError.value = '评测详情加载失败，请稍后重试。'
+  } finally {
+    loading.value = false
+  }
+}
+
+onMounted(loadDetail)
+</script>
+
+<template>
+  <PageCard class="eval-detail-page" v-loading="loading">
+    <template #header>
+      <div class="page-heading">
+        <div class="heading-copy">
+          <div class="heading-row">
+            <h1>{{ detail?.eval_task_name || '评测任务详情' }}</h1>
+            <ModelStatusTag v-if="detail" :status="detail.status" />
+          </div>
+          <dl class="task-meta">
+            <div><dt>任务 ID</dt><dd>{{ detail?.id || taskId }}</dd></div>
+            <div><dt>评测模型</dt><dd>{{ detail?.model_name || '-' }}</dd></div>
+            <div><dt>测试集</dt><dd>{{ detail?.dataset || '-' }}</dd></div>
+            <div><dt>评测指标</dt><dd>{{ detail?.metric || '-' }}</dd></div>
+          </dl>
+        </div>
+      </div>
+    </template>
+
+    <div v-if="loadError" class="load-state is-error" role="alert">
+      <i class="fa fa-exclamation-circle" aria-hidden="true" />
+      <h2>无法加载评测详情</h2>
+      <p>{{ loadError }}</p>
+      <el-button type="primary" @click="loadDetail">重新加载</el-button>
+    </div>
+
+    <template v-else-if="detail">
+      <section class="overview-grid" aria-label="评测概览">
+        <div class="overview-item score-hero">
+          <span>综合得分</span>
+          <strong>{{ overallScore }}</strong>
+          <small>大模型综合评分</small>
+        </div>
+        <div class="overview-item">
+          <span>样本通过率</span>
+          <strong>{{ detail.completed_count ? `${passRate}%` : '-' }}</strong>
+          <small>通过 {{ detail.passed_count || 0 }} / 已评测 {{ detail.completed_count }}</small>
+        </div>
+        <div class="overview-item">
+          <span>评测进度</span>
+          <strong>{{ detail.completed_count }} / {{ detail.sample_count }}</strong>
+          <el-progress :percentage="completionRate" :show-text="false" :stroke-width="5" />
+        </div>
+        <div class="overview-item overview-time">
+          <span>{{ detail.completed_time ? '完成时间' : '创建时间' }}</span>
+          <strong>{{ formatDateTime(detail.completed_time || detail.create_time) }}</strong>
+          <small>{{ detail.completed_time ? '评测任务已完成' : `当前进度 ${completionRate}%` }}</small>
+        </div>
+      </section>
+
+      <section class="overall-review" aria-labelledby="overall-review-title">
+        <div class="review-copy">
+          <div class="section-heading">
+            <div>
+              <h2 id="overall-review-title">大模型综合评价</h2>
+              <p>基于全部已评测样本生成的总体结论</p>
+            </div>
+            <el-tag v-if="detail.evaluator_model" type="primary" size="small">
+              评审模型：{{ detail.evaluator_model }}
+            </el-tag>
+          </div>
+          <p class="review-text">
+            {{ detail.overall_evaluation || (detail.status === 'running' ? '评测仍在进行，综合评价将在样本评分完成后生成。' : '暂无综合评价。') }}
+          </p>
+
+          <div class="suggestion-block">
+            <h3>改进建议</h3>
+            <ul v-if="detail.improvement_suggestions?.length" class="improvement-list">
+              <li v-for="suggestion in detail.improvement_suggestions" :key="suggestion">
+                {{ suggestion }}
+              </li>
+            </ul>
+            <p v-else class="empty-copy">暂无改进建议。</p>
+          </div>
+        </div>
+      </section>
+
+      <section v-if="detail.dimension_summary?.length" class="dimension-summary" aria-labelledby="dimension-title">
+        <div class="section-heading compact-heading">
+          <div>
+            <h2 id="dimension-title">维度表现</h2>
+            <p>查看各评测维度的得分与样本通过率</p>
+          </div>
+        </div>
+        <div class="dimension-grid">
+          <div v-for="dimension in detail.dimension_summary" :key="dimension.name" class="dimension-item">
+            <div class="dimension-label">
+              <strong>{{ dimension.name }}</strong>
+              <span>{{ formatScore(dimension.score, dimension.max_score) }}</span>
+            </div>
+            <el-progress :percentage="dimension.pass_rate" :stroke-width="7" />
+          </div>
+        </div>
+      </section>
+
+      <section class="sample-section" aria-labelledby="sample-results-title">
+        <div class="sample-toolbar">
+          <div>
+            <h2 id="sample-results-title">测试集样本评测结果</h2>
+            <p>共 {{ filteredSamples.length }} 条结果，展开行可查看评分依据与子维度分数</p>
+          </div>
+          <div class="sample-filters" aria-label="样本筛选">
+            <el-input
+              v-model="keyword"
+              clearable
+              placeholder="搜索问题、回答或评价"
+              aria-label="搜索样本"
+              @input="resetPage"
+            >
+              <template #prefix><i class="fa fa-search" aria-hidden="true" /></template>
+            </el-input>
+            <el-select
+              v-model="judgementFilter"
+              clearable
+              placeholder="全部判定"
+              aria-label="按判定筛选"
+              @change="resetPage"
+            >
+              <el-option label="正确" value="正确" />
+              <el-option label="部分正确" value="部分正确" />
+              <el-option label="错误" value="错误" />
+            </el-select>
+          </div>
+        </div>
+
+        <el-table
+          v-if="filteredSamples.length"
+          class="sample-results-table"
+          :data="paginatedSamples"
+          row-key="id"
+          table-layout="fixed"
+        >
+          <el-table-column type="expand" width="48">
+            <template #default="{ row }">
+              <div class="sample-detail-grid">
+                <div class="evaluation-reason">
+                  <span>大模型评分依据</span>
+                  <p>{{ row.evaluation_reason || '暂无评分依据。' }}</p>
+                </div>
+                <div v-if="row.error_type" class="error-type">
+                  <span>错误类型</span>
+                  <el-tag type="danger" size="small">{{ row.error_type }}</el-tag>
+                </div>
+                <div v-if="row.dimension_scores?.length" class="sample-dimensions">
+                  <span>子维度评分</span>
+                  <div class="dimension-score-list">
+                    <span v-for="item in row.dimension_scores" :key="item.name">
+                      {{ item.name }} <strong>{{ formatScore(item.score, item.max_score) }}</strong>
+                    </span>
+                  </div>
+                </div>
+              </div>
+            </template>
+          </el-table-column>
+          <el-table-column label="#" prop="index" width="56" align="center" />
+          <el-table-column label="问题 / 输入" min-width="180">
+            <template #default="{ row }"><p class="cell-copy">{{ row.input }}</p></template>
+          </el-table-column>
+          <el-table-column label="标准答案" min-width="170">
+            <template #default="{ row }"><p class="cell-copy">{{ row.reference_answer || '-' }}</p></template>
+          </el-table-column>
+          <el-table-column label="模型回答" min-width="190">
+            <template #default="{ row }"><p class="cell-copy">{{ row.model_output || '等待生成' }}</p></template>
+          </el-table-column>
+          <el-table-column label="得分" width="90" align="center">
+            <template #default="{ row }">
+              <span class="sample-score">{{ formatScore(row.score, row.max_score) }}</span>
+            </template>
+          </el-table-column>
+          <el-table-column label="判定" width="96" align="center">
+            <template #default="{ row }">
+              <el-tag :type="judgementTagType(row.judgement)" size="small">
+                {{ row.judgement || (row.status === 'pending' ? '待评测' : '-') }}
+              </el-tag>
+            </template>
+          </el-table-column>
+        </el-table>
+
+        <div v-else class="load-state sample-empty">
+          <i class="fa fa-list-alt" aria-hidden="true" />
+          <h3>{{ detail.samples.length ? '没有符合筛选条件的样本' : '暂无样本评测结果' }}</h3>
+          <p>{{ detail.status === 'running' ? '任务正在运行，结果生成后会显示在这里。' : '请调整筛选条件或稍后重试。' }}</p>
+        </div>
+
+        <el-pagination
+          v-if="filteredSamples.length > pageSize"
+          v-model:current-page="currentPage"
+          v-model:page-size="pageSize"
+          background
+          layout="total, sizes, prev, pager, next"
+          :page-sizes="[10, 20, 50]"
+          :total="filteredSamples.length"
+          aria-label="样本结果分页"
+        />
+      </section>
+    </template>
+  </PageCard>
+</template>
+
+<style scoped lang="scss">
+.eval-detail-page {
+  min-width: 0;
+}
+
+.page-heading,
+.heading-row,
+.section-heading,
+.sample-toolbar,
+.sample-filters,
+.dimension-label {
+  display: flex;
+  align-items: center;
+}
+
+.heading-copy {
+  width: 100%;
+  min-width: 0;
+}
+
+.heading-row {
+  gap: 10px;
+  flex-wrap: wrap;
+}
+
+.heading-row h1 {
+  margin: 0;
+  color: #303133;
+  font-size: 18px;
+  font-weight: 600;
+  line-height: 1.4;
+}
+
+.task-meta {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 8px 24px;
+  margin: 10px 0 0;
+}
+
+.task-meta > div {
+  display: flex;
+  gap: 6px;
+  min-width: 0;
+}
+
+.task-meta dt,
+.task-meta dd {
+  margin: 0;
+  font-size: 13px;
+}
+
+.task-meta dt {
+  color: #909399;
+}
+
+.task-meta dd {
+  color: #606266;
+  font-weight: 500;
+}
+
+.overview-grid {
+  display: grid;
+  grid-template-columns: repeat(4, minmax(0, 1fr));
+  border: 1px solid #e4e7ed;
+  border-radius: 8px;
+  overflow: hidden;
+  background: #fff;
+}
+
+.overview-item {
+  min-width: 0;
+  min-height: 92px;
+  padding: 16px 20px;
+  border-right: 1px solid #e4e7ed;
+}
+
+.overview-item:last-child {
+  border-right: 0;
+}
+
+.overview-item > span,
+.evaluation-reason > span,
+.error-type > span,
+.sample-dimensions > span {
+  display: block;
+  margin-bottom: 8px;
+  color: #606266;
+  font-size: 13px;
+  font-weight: 500;
+}
+
+.overview-item > strong {
+  display: block;
+  color: #303133;
+  font-size: 22px;
+  font-weight: 600;
+  line-height: 1.3;
+  font-variant-numeric: tabular-nums;
+}
+
+.score-hero > strong {
+  color: var(--primary-color);
+}
+
+.overview-item small {
+  display: block;
+  margin-top: 5px;
+  color: #909399;
+  font-size: 12px;
+}
+
+.overview-item :deep(.el-progress) {
+  margin-top: 8px;
+}
+
+.overview-time > strong {
+  font-size: 14px;
+  font-weight: 500;
+  line-height: 1.6;
+}
+
+.overall-review {
+  margin-top: 16px;
+  border: 1px solid #e4e7ed;
+  border-radius: 8px;
+  background: #fff;
+}
+
+.review-copy {
+  padding: 18px 20px;
+}
+
+.section-heading,
+.sample-toolbar {
+  justify-content: space-between;
+  gap: 16px;
+}
+
+.section-heading h2,
+.sample-toolbar h2 {
+  margin: 0;
+  color: #303133;
+  font-size: 16px;
+  font-weight: 600;
+}
+
+.section-heading p,
+.sample-toolbar p {
+  margin: 5px 0 0;
+  color: #909399;
+  font-size: 13px;
+}
+
+.review-text {
+  max-width: 1100px;
+  margin: 16px 0;
+  color: #606266;
+  font-size: 14px;
+  line-height: 1.75;
+}
+
+.suggestion-block {
+  padding: 14px 16px;
+  border: 1px solid #ebeef5;
+  border-radius: 6px;
+  background: #f8f9fb;
+}
+
+.suggestion-block h3 {
+  margin: 0 0 8px;
+  color: #303133;
+  font-size: 14px;
+  font-weight: 600;
+}
+
+.improvement-list {
+  display: grid;
+  grid-template-columns: repeat(2, minmax(0, 1fr));
+  gap: 6px 24px;
+  margin: 0;
+  padding-left: 18px;
+  color: #606266;
+  font-size: 13px;
+  line-height: 1.6;
+}
+
+.dimension-summary,
+.sample-section {
+  margin-top: 16px;
+  padding: 18px 20px;
+  border: 1px solid #e4e7ed;
+  border-radius: 8px;
+  background: #fff;
+}
+
+.compact-heading {
+  align-items: center;
+}
+
+.dimension-grid {
+  display: grid;
+  grid-template-columns: repeat(4, minmax(0, 1fr));
+  gap: 0;
+  margin-top: 16px;
+  border: 1px solid #ebeef5;
+  border-radius: 6px;
+  overflow: hidden;
+}
+
+.dimension-item {
+  min-width: 0;
+  padding: 14px 16px;
+  border-right: 1px solid #ebeef5;
+  background: #fafafa;
+}
+
+.dimension-item:last-child {
+  border-right: 0;
+}
+
+.dimension-label {
+  justify-content: space-between;
+  margin-bottom: 8px;
+  gap: 8px;
+  color: #606266;
+  font-size: 13px;
+}
+
+.dimension-label strong {
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
+.dimension-label span {
+  flex: 0 0 auto;
+  color: #303133;
+  font-weight: 600;
+  font-variant-numeric: tabular-nums;
+}
+
+.sample-toolbar {
+  align-items: center;
+  margin-bottom: 14px;
+}
+
+.sample-filters {
+  flex: 0 1 450px;
+  justify-content: flex-end;
+  gap: 10px;
+}
+
+.sample-filters :deep(.el-input) {
+  width: min(260px, 100%);
+}
+
+.sample-filters :deep(.el-select) {
+  width: 130px;
+}
+
+.sample-results-table {
+  width: 100%;
+  border: 1px solid #ebeef5;
+  border-radius: 6px;
+}
+
+.cell-copy {
+  display: -webkit-box;
+  overflow: hidden;
+  margin: 0;
+  color: #606266;
+  font-size: 13px;
+  line-height: 1.55;
+  -webkit-box-orient: vertical;
+  -webkit-line-clamp: 2;
+}
+
+.sample-score {
+  color: #303133;
+  font-weight: 600;
+  font-variant-numeric: tabular-nums;
+}
+
+.sample-detail-grid {
+  display: grid;
+  grid-template-columns: minmax(140px, 0.35fr) minmax(320px, 1fr);
+  gap: 14px 24px;
+  padding: 16px 48px;
+  background: #f7f8fa;
+}
+
+.evaluation-reason {
+  grid-column: 1 / -1;
+}
+
+.evaluation-reason p {
+  margin: 0;
+  color: #606266;
+  font-size: 13px;
+  line-height: 1.7;
+}
+
+.dimension-score-list {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 8px;
+}
+
+.dimension-score-list > span {
+  color: #606266;
+  font-size: 12px;
+}
+
+.dimension-score-list > span:not(:last-child)::after {
+  margin-left: 8px;
+  color: #c0c4cc;
+  content: '/';
+}
+
+.dimension-score-list strong {
+  margin-left: 4px;
+  color: #303133;
+  font-variant-numeric: tabular-nums;
+}
+
+.load-state {
+  min-height: 280px;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  justify-content: center;
+  padding: 32px;
+  color: #909399;
+  text-align: center;
+}
+
+.load-state > i {
+  margin-bottom: 12px;
+  color: #c0c4cc;
+  font-size: 30px;
+}
+
+.load-state h2,
+.load-state h3 {
+  margin: 0;
+  color: #303133;
+}
+
+.load-state p,
+.empty-copy {
+  margin: 8px 0 16px;
+  color: #909399;
+  font-size: 13px;
+}
+
+.load-state.is-error > i {
+  color: #ef4444;
+}
+
+.sample-empty {
+  min-height: 210px;
+  border: 1px dashed #dcdfe6;
+  border-radius: 6px;
+  background: #fafafa;
+}
+
+@media (max-width: 1100px) {
+  .overview-grid {
+    grid-template-columns: repeat(2, minmax(0, 1fr));
+  }
+
+  .overview-item:nth-child(2) {
+    border-right: 0;
+  }
+
+  .overview-item:nth-child(-n + 2) {
+    border-bottom: 1px solid #e4e7ed;
+  }
+
+  .dimension-grid {
+    grid-template-columns: repeat(2, minmax(0, 1fr));
+  }
+
+  .dimension-item:nth-child(2) {
+    border-right: 0;
+  }
+
+  .dimension-item:nth-child(-n + 2) {
+    border-bottom: 1px solid #ebeef5;
+  }
+
+  .sample-toolbar {
+    align-items: stretch;
+    flex-direction: column;
+  }
+
+  .sample-filters {
+    flex: 1;
+    justify-content: flex-start;
+  }
+
+  .sample-detail-grid {
+    padding: 16px 24px;
+  }
+}
+
+@media (max-width: 720px) {
+  .task-meta {
+    flex-direction: column;
+    gap: 5px;
+  }
+
+  .overview-grid {
+    grid-template-columns: repeat(2, minmax(0, 1fr));
+  }
+
+  .overview-item {
+    padding: 14px;
+  }
+
+  .review-copy,
+  .dimension-summary,
+  .sample-section {
+    padding: 16px;
+  }
+
+  .dimension-grid,
+  .improvement-list,
+  .sample-detail-grid {
+    grid-template-columns: minmax(0, 1fr);
+  }
+
+  .dimension-item,
+  .dimension-item:nth-child(2) {
+    border-right: 0;
+    border-bottom: 1px solid #ebeef5;
+  }
+
+  .dimension-item:last-child {
+    border-bottom: 0;
+  }
+
+  .sample-filters {
+    flex-direction: column;
+    align-items: stretch;
+  }
+
+  .sample-filters :deep(.el-input),
+  .sample-filters :deep(.el-select) {
+    width: 100%;
+  }
+
+  .evaluation-reason {
+    grid-column: auto;
+  }
+}
+</style>

frontend/src/views/eval/EvalView.vue

@@ -68,15 +68,15 @@ function editDimension(row: any) {
 }
 
 function handleCreateClick() {
-  if (activeTab.value === 'tasks') {
   router.push('/model-eval/create')
-  } else if (activeTab.value === 'dimensions') {
+}
-    router.push('/model-eval/dimension/create')
+
-  }
+function isActiveTab(tab: string) {
+  return activeTab.value === tab
 }
 
 function handleViewDetail(row: any) {
-  ElMessage.info('详情功能开发中')
+  router.push({ name: 'model-eval-detail', params: { id: row.id } })
 }
 
 onMounted(() => {
@@ -101,9 +101,9 @@ onMounted(() => {
     >
       <template #title>
         <div class="capsule-tabs">
-          <button class="capsule-tab-item" :class="{ active: activeTab === 'tasks' }" @click="activeTab = 'tasks'">评测任务</button>
+          <button class="capsule-tab-item" :class="{ active: isActiveTab('tasks') }" @click="activeTab = 'tasks'">评测任务</button>
-          <button class="capsule-tab-item" :class="{ active: activeTab === 'leaderboard' }" @click="activeTab = 'leaderboard'">排行榜</button>
+          <button class="capsule-tab-item" :class="{ active: isActiveTab('leaderboard') }" @click="activeTab = 'leaderboard'">排行榜</button>
-          <button class="capsule-tab-item" :class="{ active: activeTab === 'dimensions' }" @click="activeTab = 'dimensions'">评测维度</button>
+          <button class="capsule-tab-item" :class="{ active: isActiveTab('dimensions') }" @click="activeTab = 'dimensions'">评测维度</button>
         </div>
       </template>
             <template #columns>
@@ -134,9 +134,9 @@ onMounted(() => {
     >
       <template #title>
         <div class="capsule-tabs">
-          <button class="capsule-tab-item" :class="{ active: activeTab === 'tasks' }" @click="activeTab = 'tasks'">评测任务</button>
+          <button class="capsule-tab-item" :class="{ active: isActiveTab('tasks') }" @click="activeTab = 'tasks'">评测任务</button>
-          <button class="capsule-tab-item" :class="{ active: activeTab === 'leaderboard' }" @click="activeTab = 'leaderboard'">排行榜</button>
+          <button class="capsule-tab-item" :class="{ active: isActiveTab('leaderboard') }" @click="activeTab = 'leaderboard'">排行榜</button>
-          <button class="capsule-tab-item" :class="{ active: activeTab === 'dimensions' }" @click="activeTab = 'dimensions'">评测维度</button>
+          <button class="capsule-tab-item" :class="{ active: isActiveTab('dimensions') }" @click="activeTab = 'dimensions'">评测维度</button>
         </div>
       </template>
       <template #columns>
@@ -153,16 +153,14 @@ onMounted(() => {
       :data="dimensionList"
       :loading="dimLoading"
       :delete-fn="handleDeleteDimension"
-      create-text="添加维度"
-      @create="handleCreateClick"
       row-key="id"
       @refresh="loadDimensions"
     >
       <template #title>
         <div class="capsule-tabs">
-          <button class="capsule-tab-item" :class="{ active: activeTab === 'tasks' }" @click="activeTab = 'tasks'">评测任务</button>
+          <button class="capsule-tab-item" :class="{ active: isActiveTab('tasks') }" @click="activeTab = 'tasks'">评测任务</button>
-          <button class="capsule-tab-item" :class="{ active: activeTab === 'leaderboard' }" @click="activeTab = 'leaderboard'">排行榜</button>
+          <button class="capsule-tab-item" :class="{ active: isActiveTab('leaderboard') }" @click="activeTab = 'leaderboard'">排行榜</button>
-          <button class="capsule-tab-item" :class="{ active: activeTab === 'dimensions' }" @click="activeTab = 'dimensions'">评测维度</button>
+          <button class="capsule-tab-item" :class="{ active: isActiveTab('dimensions') }" @click="activeTab = 'dimensions'">评测维度</button>
         </div>
       </template>
             <template #columns>

frontend/src/views/eval/create/DimensionFormFields.vue

@@ -0,0 +1,168 @@
+<script setup lang="ts">
+import { computed, watch } from 'vue'
+import { MdEditor } from 'md-editor-v3'
+import 'md-editor-v3/lib/style.css'
+import { EVAL_METHODS, EVAL_METHOD_PROMPTS } from '@/constants/dimension'
+import { DIMENSION_TYPE_MAP } from '@/constants'
+import type { DimensionType, ModelItem } from '@/types'
+
+export interface DimensionFormDraft {
+  name: string
+  type: DimensionType | ''
+  description: string
+  eval_model: string
+  eval_method: string | string[]
+  eval_prompt: string
+  is_active: boolean
+  is_default: boolean
+  bleu_n: number
+  output_precision: number
+  score_min: number
+  score_max: number
+  pass_threshold: number
+}
+
+defineProps<{ evalModels: ModelItem[] }>()
+
+const form = defineModel<DimensionFormDraft>({ required: true })
+
+const typeOptions = Object.entries(DIMENSION_TYPE_MAP).map(([value, label]) => ({ value, label }))
+const currentMethods = computed(() => (form.value.type ? EVAL_METHODS[form.value.type] || [] : []))
+
+/**
+ * 指标类型变化时初始化对应的评估方式和 Prompt。
+ * 同步执行可保证编辑页批量回填时，接口返回值最终覆盖默认值。
+ */
+watch(
+  () => form.value.type,
+  (type) => {
+    if (!type) return
+    const firstMethod = EVAL_METHODS[type]?.[0]?.value || ''
+    form.value.eval_method = type === 'text_similarity' ? [firstMethod] : firstMethod
+    if (type !== 'text_similarity' && EVAL_METHOD_PROMPTS[firstMethod]) {
+      form.value.eval_prompt = EVAL_METHOD_PROMPTS[firstMethod]
+    }
+  },
+  { flush: 'sync' },
+)
+
+/** 评估方式变化时应用对应的预置 Prompt。 */
+watch(
+  () => form.value.eval_method,
+  (method) => {
+    if (form.value.type === 'text_similarity' || !method) return
+    const selectedMethod = Array.isArray(method) ? method[0] : method
+    if (selectedMethod && EVAL_METHOD_PROMPTS[selectedMethod]) {
+      form.value.eval_prompt = EVAL_METHOD_PROMPTS[selectedMethod]
+    }
+  },
+  { flush: 'sync' },
+)
+</script>
+
+<template>
+  <el-form-item label="维度名称" prop="name">
+    <el-input v-model="form.name" placeholder="请输入维度名称" maxlength="50" show-word-limit />
+  </el-form-item>
+
+  <el-form-item label="指标类型" prop="type">
+    <el-select v-model="form.type" placeholder="请选择指标类型" style="width: 100%">
+      <el-option v-for="option in typeOptions" :key="option.value" :label="option.label" :value="option.value" />
+    </el-select>
+  </el-form-item>
+
+  <el-form-item label="描述">
+    <el-input v-model="form.description" type="textarea" :rows="2" maxlength="200" show-word-limit />
+  </el-form-item>
+
+  <template v-if="form.type === 'classification' || form.type === 'metric'">
+    <el-form-item label="选择大模型" prop="eval_model">
+      <el-select v-model="form.eval_model" placeholder="请选择评测模型" filterable style="width: 100%">
+        <el-option v-for="model in evalModels" :key="model.id" :label="model.name" :value="model.name" />
+      </el-select>
+    </el-form-item>
+
+    <el-form-item label="评估方式" prop="eval_method">
+      <el-radio-group v-model="(form.eval_method as string)">
+        <el-radio v-for="method in currentMethods" :key="method.value" :value="method.value">
+          {{ method.name }}
+          <span class="method-desc">{{ method.desc }}</span>
+        </el-radio>
+      </el-radio-group>
+    </el-form-item>
+
+    <el-form-item label="评估 Prompt" prop="eval_prompt">
+      <MdEditor
+        v-model="form.eval_prompt"
+        language="zh-CN"
+        :toolbars-exclude="['github', 'save']"
+        style="height: 320px"
+      />
+    </el-form-item>
+  </template>
+
+  <template v-if="form.type === 'text_similarity'">
+    <el-form-item label="评估方式" prop="eval_method">
+      <el-checkbox-group v-model="(form.eval_method as string[])">
+        <el-checkbox v-for="method in currentMethods" :key="method.value" :value="method.value">
+          {{ method.name }}
+          <span class="method-desc">{{ method.desc }}</span>
+        </el-checkbox>
+      </el-checkbox-group>
+    </el-form-item>
+
+    <el-form-item label="BLEU n-gram">
+      <el-select v-model="form.bleu_n" style="width: 160px">
+        <el-option :value="1" label="1-gram" />
+        <el-option :value="2" label="2-gram" />
+        <el-option :value="3" label="3-gram" />
+        <el-option :value="4" label="4-gram" />
+      </el-select>
+    </el-form-item>
+
+    <el-form-item label="输出精度">
+      <el-select v-model="form.output_precision" style="width: 160px">
+        <el-option :value="1" label="1 位小数" />
+        <el-option :value="2" label="2 位小数" />
+        <el-option :value="3" label="3 位小数" />
+        <el-option :value="4" label="4 位小数" />
+      </el-select>
+    </el-form-item>
+  </template>
+
+  <template v-if="form.type === 'metric'">
+    <el-form-item label="评分最小值">
+      <el-input-number v-model="form.score_min" :min="0" :step="1" style="width: 160px" />
+    </el-form-item>
+    <el-form-item label="评分最大值">
+      <el-input-number v-model="form.score_max" :min="0" :step="1" style="width: 160px" />
+    </el-form-item>
+    <el-form-item label="通过阈值">
+      <div style="width: 100%">
+        <el-slider
+          v-model="form.pass_threshold"
+          :min="form.score_min"
+          :max="form.score_max"
+          :step="0.5"
+          show-input
+          :show-input-controls="false"
+        />
+      </div>
+    </el-form-item>
+  </template>
+
+  <el-form-item label="启用该维度">
+    <el-switch v-model="form.is_active" />
+  </el-form-item>
+  <el-form-item label="设为默认">
+    <el-switch v-model="form.is_default" />
+  </el-form-item>
+</template>
+
+<style scoped>
+.method-desc {
+  color: #909399;
+  font-size: 12px;
+  margin-left: 6px;
+}
+</style>

frontend/src/views/eval/create/EvalRuleSetupStep.vue

@@ -0,0 +1,159 @@
+<script setup lang="ts">
+import { computed, ref } from 'vue'
+import type { FormInstance, FormRules } from 'element-plus'
+import DimensionFormFields, { type DimensionFormDraft } from './DimensionFormFields.vue'
+import type { Dimension, ModelItem } from '@/types'
+
+export type EvalRuleMode = 'existing' | 'new' | 'baseline'
+
+export interface EvalRuleSetupDraft {
+  ruleMode: EvalRuleMode
+  dimension_id: string | number
+  newDimension: DimensionFormDraft
+}
+
+const props = defineProps<{
+  dimensions: Dimension[]
+  evalModels: ModelItem[]
+  loading?: boolean
+  disabled?: boolean
+}>()
+
+const form = defineModel<EvalRuleSetupDraft>({ required: true })
+const existingFormRef = ref<FormInstance>()
+const newFormRef = ref<FormInstance>()
+
+const existingRules: FormRules = {
+  dimension_id: [{ required: true, message: '请选择评测维度', trigger: 'change' }],
+}
+
+const dimensionRules: FormRules<DimensionFormDraft> = {
+  name: [
+    { required: true, message: '请输入维度名称', trigger: 'blur' },
+    { max: 50, message: '维度名称不超过 50 个字符', trigger: 'blur' },
+  ],
+  type: [{ required: true, message: '请选择指标类型', trigger: 'change' }],
+  eval_model: [{ required: true, message: '请选择大模型', trigger: 'change' }],
+  eval_method: [{ required: true, message: '请选择评估方式', trigger: 'change' }],
+  eval_prompt: [{ required: true, message: '请填写评估 Prompt', trigger: 'blur' }],
+}
+
+const selectedDimension = computed(() =>
+  props.dimensions.find((dimension) => String(dimension.id) === String(form.value.dimension_id)),
+)
+
+async function validate() {
+  if (form.value.ruleMode === 'baseline') return true
+  const target = form.value.ruleMode === 'existing' ? existingFormRef.value : newFormRef.value
+  if (!target) return false
+  try {
+    await target.validate()
+    return true
+  } catch {
+    return false
+  }
+}
+
+defineExpose({ validate })
+</script>
+
+<template>
+  <div class="rule-step">
+    <div class="rule-mode-label">评测规则来源</div>
+    <el-radio-group v-model="form.ruleMode" :disabled="disabled" class="rule-mode-group">
+      <el-radio-button value="existing">选择已有维度</el-radio-button>
+      <el-radio-button value="new">新建评测维度</el-radio-button>
+      <el-radio-button value="baseline">系统基线规则</el-radio-button>
+    </el-radio-group>
+
+    <el-form
+      v-if="form.ruleMode === 'existing'"
+      ref="existingFormRef"
+      :model="form"
+      :rules="existingRules"
+      label-width="120px"
+      class="step-form"
+      :disabled="disabled"
+    >
+      <el-form-item label="评测维度" prop="dimension_id">
+        <el-select
+          v-model="form.dimension_id"
+          placeholder="请选择评测维度"
+          filterable
+          style="width: 100%"
+          :loading="loading"
+        >
+          <el-option
+            v-for="dimension in dimensions"
+            :key="dimension.id"
+            :label="dimension.name"
+            :value="dimension.id"
+          />
+        </el-select>
+      </el-form-item>
+
+      <el-form-item v-if="selectedDimension" label="维度详情">
+        <el-descriptions :column="1" border size="small" style="width: 100%">
+          <el-descriptions-item label="指标类型">{{ selectedDimension.type || '-' }}</el-descriptions-item>
+          <el-descriptions-item label="评估模型">{{ selectedDimension.eval_model || '-' }}</el-descriptions-item>
+          <el-descriptions-item label="评估方法">
+            {{
+              Array.isArray(selectedDimension.eval_method)
+                ? selectedDimension.eval_method.join('、')
+                : selectedDimension.eval_method || '-'
+            }}
+          </el-descriptions-item>
+          <el-descriptions-item v-if="selectedDimension.description" label="说明">
+            {{ selectedDimension.description }}
+          </el-descriptions-item>
+        </el-descriptions>
+      </el-form-item>
+    </el-form>
+
+    <el-form
+      v-else-if="form.ruleMode === 'new'"
+      ref="newFormRef"
+      :model="form.newDimension"
+      :rules="dimensionRules"
+      label-width="130px"
+      class="step-form step-form-wide"
+      :disabled="disabled"
+    >
+      <DimensionFormFields v-model="form.newDimension" :eval-models="evalModels" />
+    </el-form>
+
+    <el-alert
+      v-else
+      title="系统将使用内置基线规则进行评测，无需额外创建或选择评测维度。"
+      type="info"
+      :closable="false"
+      show-icon
+      class="baseline-alert"
+    />
+  </div>
+</template>
+
+<style scoped>
+.rule-mode-label {
+  margin-bottom: 10px;
+  color: #606266;
+  font-size: 14px;
+  font-weight: 500;
+}
+
+.rule-mode-group {
+  margin-bottom: 28px;
+}
+
+.step-form {
+  max-width: 760px;
+}
+
+.step-form-wide {
+  max-width: 920px;
+}
+
+.baseline-alert {
+  max-width: 760px;
+}
+</style>

frontend/src/views/eval/create/EvalTaskSetupStep.vue

@@ -0,0 +1,162 @@
+<script setup lang="ts">
+import { ref, watch } from 'vue'
+import type { FormInstance, FormRules } from 'element-plus'
+import type { DatasetItem, GpuInfo, TrainedModel } from '@/types'
+
+export interface EvalTaskSetupDraft {
+  eval_task_name: string
+  model_id: string | number
+  gpu_id: string | number
+  data_source: 'dataset' | 'inference'
+  dataset_id: string | number
+  leaderboard: boolean
+}
+
+defineProps<{
+  trainedModels: TrainedModel[]
+  evalDatasets: DatasetItem[]
+  gpus: GpuInfo[]
+  loading?: boolean
+  disabled?: boolean
+}>()
+
+const form = defineModel<EvalTaskSetupDraft>({ required: true })
+const formRef = ref<FormInstance>()
+
+const rules: FormRules<EvalTaskSetupDraft> = {
+  eval_task_name: [
+    { required: true, message: '请输入任务名称', trigger: 'blur' },
+    { max: 50, message: '任务名称不超过 50 个字符', trigger: 'blur' },
+  ],
+  model_id: [{ required: true, message: '请选择评测模型', trigger: 'change' }],
+  gpu_id: [{ required: true, message: '请选择 GPU', trigger: 'change' }],
+  dataset_id: [
+    {
+      validator: (_rule, value, callback) => {
+        if (form.value.data_source === 'dataset' && (value === '' || value === null || value === undefined)) {
+          callback(new Error('请选择评测数据集'))
+          return
+        }
+        callback()
+      },
+      trigger: 'change',
+    },
+  ],
+}
+
+watch(
+  () => form.value.data_source,
+  (source) => {
+    if (source !== 'dataset') {
+      form.value.dataset_id = ''
+      formRef.value?.clearValidate('dataset_id')
+    }
+  },
+)
+
+async function validate() {
+  if (!formRef.value) return false
+  try {
+    await formRef.value.validate()
+    return true
+  } catch {
+    return false
+  }
+}
+
+defineExpose({ validate })
+</script>
+
+<template>
+  <el-form
+    ref="formRef"
+    :model="form"
+    :rules="rules"
+    label-width="120px"
+    class="step-form"
+    :disabled="disabled"
+  >
+    <el-form-item label="任务名称" prop="eval_task_name">
+      <el-input v-model="form.eval_task_name" placeholder="请输入任务名称" maxlength="50" show-word-limit />
+    </el-form-item>
+
+    <el-form-item label="评测模型" prop="model_id">
+      <el-select
+        v-model="form.model_id"
+        placeholder="请选择已训练且已合并的模型"
+        filterable
+        style="width: 100%"
+        :loading="loading"
+      >
+        <el-option
+          v-for="model in trainedModels"
+          :key="model.id"
+          :label="model.name"
+          :value="model.id"
+          :disabled="!model.merged"
+        />
+      </el-select>
+      <div class="field-tip">未完成合并的模型不可用于评测。</div>
+    </el-form-item>
+
+    <el-form-item label="选择 GPU" prop="gpu_id">
+      <el-select v-model="form.gpu_id" placeholder="请选择 GPU" style="width: 100%" :loading="loading">
+        <el-option
+          v-for="(gpu, index) in gpus"
+          :key="index"
+          :label="`${gpu.name} (GPU ${index})`"
+          :value="index"
+        />
+      </el-select>
+    </el-form-item>
+
+    <el-form-item label="数据来源">
+      <el-radio-group v-model="form.data_source">
+        <el-radio value="dataset">评测数据集</el-radio>
+        <el-radio value="inference">推理结果集</el-radio>
+      </el-radio-group>
+    </el-form-item>
+
+    <el-form-item v-if="form.data_source === 'dataset'" label="评测数据集" prop="dataset_id">
+      <el-select
+        v-model="form.dataset_id"
+        placeholder="请选择评测数据集"
+        filterable
+        style="width: 100%"
+        :loading="loading"
+      >
+        <el-option v-for="dataset in evalDatasets" :key="dataset.id" :label="dataset.name" :value="dataset.id" />
+      </el-select>
+    </el-form-item>
+
+    <el-alert
+      v-else
+      title="将使用该模型最近一次可用的推理结果集。"
+      type="info"
+      :closable="false"
+      show-icon
+      class="source-alert"
+    />
+
+    <el-form-item label="加入排行榜">
+      <el-switch v-model="form.leaderboard" />
+    </el-form-item>
+  </el-form>
+</template>
+
+<style scoped>
+.step-form {
+  max-width: 760px;
+}
+
+.field-tip {
+  color: #909399;
+  font-size: 12px;
+  line-height: 20px;
+}
+
+.source-alert {
+  margin: -4px 0 20px 120px;
+  max-width: 640px;
+}
+</style>