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

@@ -11,6 +11,8 @@ const [
   createSource,
   taskStepSource,
   ruleStepSource,
+  basicMetricStepSource,
+  startStepSource,
   dimensionFieldsSource,
   listSource,
   routerSource,
@@ -19,6 +21,8 @@ const [
   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, 'create/BasicMetricSetupStep.vue'), 'utf8'),
+  readFile(path.join(evalViewDir, 'create/StartEvalStep.vue'), 'utf8'),
   readFile(path.join(evalViewDir, 'create/DimensionFormFields.vue'), 'utf8'),
   readFile(path.join(evalViewDir, 'EvalView.vue'), 'utf8'),
   readFile(path.join(sourceRoot, 'router/index.ts'), 'utf8'),
@@ -74,21 +78,43 @@ assert.match(
 
 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, /BasicMetricSetupStep/, '向导缺少基础评测指标子组件')
+assert.match(createSource, /StartEvalStep/, '向导缺少开始评测确认子组件')
 assert.match(
   createSource,
-  /任务配置[\s\S]*?评测规则/,
+  /任务配置[\s\S]*?大模型评测指标[\s\S]*?基础评测指标[\s\S]*?开始评测/,
-  '评测创建向导必须按“任务配置、评测规则”定义两个步骤',
+  '评测创建向导必须按“任务配置、大模型评测指标、基础评测指标、开始评测”定义四个步骤',
 )
 assert.match(
   ruleStepSource,
   /:show-description=["']false["']/,
-  '新建评测任务第二步不应显示重复的规则描述输入框',
+  '大模型评测指标步骤不应显示重复的规则描述输入框',
+)
+assert.match(
+  ruleStepSource,
+  /:show-status-settings=["']false["']/,
+  '大模型评测指标步骤不应显示维度管理状态字段',
+)
+assert.match(
+  ruleStepSource,
+  /LLM_METRIC_TYPES[^=]*=\s*\[['"]classification['"],\s*['"]metric['"]\][\s\S]*?:allowed-types=["']LLM_METRIC_TYPES["']/,
+  '大模型评测指标步骤只应提供依赖大模型的分类与评分指标',
 )
 assert.match(
   dimensionFieldsSource,
   /v-if=["']showDescription["'][^>]*label=["']描述["']/,
   '共享维度字段必须支持按场景隐藏描述输入框',
 )
+assert.match(
+  dimensionFieldsSource,
+  /score_min[\s\S]*?score_max[\s\S]*?pass_threshold[\s\S]*?Math\.min[\s\S]*?Math\.max/,
+  '指标型评分区间变化时必须把通过阈值约束在合法范围内',
+)
+assert.match(
+  dimensionFieldsSource,
+  /prop=["']score_min["'][\s\S]*?:max=["'][^"']*score_max[^"']*["'][\s\S]*?prop=["']score_max["'][\s\S]*?:min=["'][^"']*score_min[^"']*["']/,
+  '评分最小值和最大值输入必须具备交叉边界约束',
+)
 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')
@@ -102,8 +128,18 @@ assert.match(
 )
 assert.match(
   createSource,
-  /<EvalRuleSetupStep[\s\S]*?v-else/,
+  /<EvalRuleSetupStep[\s\S]*?currentStep[^"']*1/,
-  '评测规则子组件没有绑定第二步',
+  '大模型评测指标子组件没有绑定第二步',
+)
+assert.match(
+  createSource,
+  /<BasicMetricSetupStep[\s\S]*?currentStep[^"']*2/,
+  '基础评测指标子组件没有绑定第三步',
+)
+assert.match(
+  createSource,
+  /<StartEvalStep[\s\S]*?v-else/,
+  '开始评测确认子组件没有绑定第四步',
 )
 
 assert.match(taskStepSource, /prop=["']dataset_id["']/, '数据集字段缺少表单校验标识')
@@ -113,16 +149,15 @@ assert.match(
   '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}`,
+  basicMetricStepSource,
-  /已有维度|existing/,
+  /BLEU[\s\S]*?ROUGE[\s\S]*?(?:Cosine|余弦)/,
-  '评测规则步骤缺少使用已有维度的分支',
+  '基础评测指标步骤必须提供 BLEU、ROUGE 与余弦相似度配置',
 )
 assert.match(
-  `${createSource}\n${ruleStepSource}`,
+  basicMetricStepSource,
-  /新建(?:评测)?(?:维度|规则)|create/,
+  /bleu_enabled[\s\S]*?rouge_enabled[\s\S]*?cosine_enabled/,
-  '评测规则步骤缺少新建维度的分支',
+  '基础评测指标必须支持分别决定是否启用',
 )
 
 assert.match(
@@ -137,17 +172,18 @@ assert.match(
 )
 assert.match(
   createSource,
-  /eval_type[\s\S]*?baseline[\s\S]*?(?:dimension_id|createDimension)/,
+  /basic_metrics[\s\S]*?bleu[\s\S]*?rouge[\s\S]*?cosine/,
-  '基线评测必须有明确的不创建自定义维度分支',
+  '启动评测时必须提交基础评测指标配置',
 )
 assert.match(
-  createSource,
+  startStepSource,
-  /ruleMode[\s\S]*?['"]existing['"]/,
+  /任务信息[\s\S]*?大模型评测指标[\s\S]*?基础评测指标/,
-  '已有维度模式必须有独立的提交分支',
+  '开始评测步骤必须展示前三步配置摘要',
 )
 
 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, />\s*开始评测\s*</, '最终提交按钮必须命名为“开始评测”')
 assert.match(createSource, /class=["'][^"']*wizard-footer/, '评测向导缺少统一底部操作栏')
 assert.match(
   createSource,

frontend/src/api/modules/eval.ts

@@ -1,5 +1,5 @@
 import { get, post, put, del } from '../request'
-import type { EvalTask, EvalTaskDetail, Dimension } from '@/types'
+import type { EvalTask, EvalTaskDetail, Dimension, StartEvalPayload } from '@/types'
 
 /** 评测任务列表 */
 export const getEvalList = () => get<EvalTask[]>('/model-eval')
@@ -11,7 +11,7 @@ export const getEvalDetail = (id: string | number) => get<EvalTaskDetail>(`/mode
 export const deleteEval = (id: string | number) => del(`/model-eval/${id}`)
 
 /** 启动评测 */
-export const startEval = (data: any) => post('/model-eval/start', data)
+export const startEval = (data: StartEvalPayload) => post('/model-eval/start', data)
 
 /** 评测维度列表 */
 export const getDimensionList = () => get<Dimension[]>('/dimension')

frontend/src/types/index.ts

@@ -224,6 +224,34 @@ export interface EvalTask {
   create_time?: string
 }
 
+/** 启动评测时提交的可选基础指标配置。 */
+export interface BasicEvalMetricsConfig {
+  bleu: {
+    enabled: boolean
+    ngram: number
+  }
+  rouge: {
+    enabled: boolean
+    methods: string[]
+  }
+  cosine: {
+    enabled: boolean
+  }
+  output_precision: number
+}
+
+export interface StartEvalPayload {
+  eval_task_name: string
+  eval_type: EvalType
+  model_id: string | number
+  gpu_id: string | number
+  dataset_id: string | number
+  dimension_id: string | number
+  data_source: 'dataset' | 'inference'
+  leaderboard: boolean
+  basic_metrics: BasicEvalMetricsConfig
+}
+
 /** 单个测试样本的评测结果 */
 export interface EvalSampleResult {
   id: number | string

frontend/src/views/eval/EvalCreateView.vue

@@ -5,7 +5,9 @@ import { ElMessage } from 'element-plus'
 import PageCard from '@/components/PageCard.vue'
 import EvalTaskSetupStep, { type EvalTaskSetupDraft } from './create/EvalTaskSetupStep.vue'
 import EvalRuleSetupStep, { type EvalRuleSetupDraft } from './create/EvalRuleSetupStep.vue'
-import { createDimension, getDimensionList, startEval } from '@/api/modules/eval'
+import BasicMetricSetupStep, { type BasicMetricSetupDraft } from './create/BasicMetricSetupStep.vue'
+import StartEvalStep from './create/StartEvalStep.vue'
+import { createDimension, startEval } from '@/api/modules/eval'
 import { getTrainedModels, getModelList } from '@/api/modules/model'
 import { getDatasetList } from '@/api/modules/dataset'
 import { getSystemInfo } from '@/api/modules/system'
@@ -19,15 +21,17 @@ const submitting = ref(false)
 const currentStep = ref(0)
 const taskStepRef = ref<StepExposed>()
 const ruleStepRef = ref<StepExposed>()
+const basicMetricStepRef = ref<StepExposed>()
 
 const WIZARD_STEPS = [
   { title: '任务配置', description: '选择模型、算力与评测数据' },
-  { title: '评测规则', description: '选择或创建评测维度' },
+  { title: '大模型评测指标', description: '配置评测模型、方式与标准' },
+  { title: '基础评测指标', description: '选择 BLEU、ROUGE 等参考指标' },
+  { title: '开始评测', description: '确认配置并启动评测任务' },
 ] as const
 
 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>('')
@@ -42,26 +46,31 @@ const taskForm = ref<EvalTaskSetupDraft>({
 })
 
 const ruleForm = ref<EvalRuleSetupDraft>({
-  ruleMode: 'new',
+  type: '',
-  dimension_id: '',
+  description: '',
-  newDimension: {
+  eval_model: '',
-    type: '',
+  eval_method: '',
-    description: '',
+  eval_prompt: '',
-    eval_model: '',
+  is_active: true,
-    eval_method: '',
+  is_default: false,
-    eval_prompt: '',
+  bleu_n: 1,
-    is_active: true,
+  output_precision: 3,
-    is_default: false,
+  score_min: 0,
-    bleu_n: 1,
+  score_max: 5,
-    output_precision: 3,
+  pass_threshold: 3,
-    score_min: 0,
+})
-    score_max: 5,
+
-    pass_threshold: 3,
+const basicMetricForm = ref<BasicMetricSetupDraft>({
-  },
+  bleu_enabled: false,
+  bleu_n: 4,
+  rouge_enabled: false,
+  rouge_methods: ['rouge_1', 'rouge_2'],
+  cosine_enabled: false,
+  output_precision: 3,
 })
 
 watch(
-  () => ruleForm.value.newDimension,
+  ruleForm,
   () => {
     createdDimensionId.value = ''
   },
@@ -72,20 +81,18 @@ async function loadData() {
   loading.value = true
   const results = await Promise.allSettled([
     getTrainedModels(),
-    getDimensionList(),
     getDatasetList(),
     getSystemInfo(),
     getModelList(),
   ])
 
   if (results[0].status === 'fulfilled') trainedModels.value = results[0].value?.models || []
-  if (results[1].status === 'fulfilled') dimensions.value = results[1].value || []
+  if (results[1].status === 'fulfilled') {
-  if (results[2].status === 'fulfilled') {
+    evalDatasets.value = (results[1].value || []).filter((dataset) => dataset.type === 'eval')
-    evalDatasets.value = (results[2].value || []).filter((dataset) => dataset.type === 'eval')
   }
-  if (results[3].status === 'fulfilled') gpus.value = results[3].value?.gpu || []
+  if (results[2].status === 'fulfilled') gpus.value = results[2].value?.gpu || []
-  if (results[4].status === 'fulfilled') {
+  if (results[3].status === 'fulfilled') {
-    evalModels.value = (results[4].value || []).filter((model) => model.purpose === 'evaluation')
+    evalModels.value = (results[3].value || []).filter((model) => model.purpose === 'evaluation')
   }
 
   const failedCount = results.filter((result) => result.status === 'rejected').length
@@ -95,25 +102,19 @@ async function loadData() {
   loading.value = false
 }
 
-
-
 function buildDimensionPayload() {
-  const draft = ruleForm.value.newDimension
+  const draft = ruleForm.value
   const payload: Partial<Dimension> = {
     name: `Custom_Dim_${Date.now()}`,
     type: draft.type,
     description: draft.description,
-    eval_model: draft.type === 'text_similarity' ? undefined : draft.eval_model,
+    eval_model: draft.eval_model,
     eval_method: draft.eval_method,
-    eval_prompt: draft.type === 'text_similarity' ? undefined : draft.eval_prompt,
+    eval_prompt: 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
@@ -123,8 +124,6 @@ function buildDimensionPayload() {
 }
 
 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())
@@ -137,24 +136,32 @@ async function resolveDimensionId() {
 
 async function handleSubmit() {
   if (loading.value || submitting.value) return
-  const ruleValid = await ruleStepRef.value?.validate()
-  if (!ruleValid) {
-    ElMessage.warning('请检查并完善评测规则')
-    return
-  }
-
   submitting.value = true
   try {
     const dimensionId = await resolveDimensionId()
     await startEval({
       eval_task_name: taskForm.value.eval_task_name,
-      eval_type: ruleForm.value.ruleMode === 'baseline' ? 'baseline' : 'custom',
+      eval_type: 'custom',
       model_id: taskForm.value.model_id,
       gpu_id: taskForm.value.gpu_id,
       dataset_id: taskForm.value.data_source === 'dataset' ? taskForm.value.dataset_id : '',
       dimension_id: dimensionId,
       data_source: taskForm.value.data_source,
       leaderboard: taskForm.value.leaderboard,
+      basic_metrics: {
+        bleu: {
+          enabled: basicMetricForm.value.bleu_enabled,
+          ngram: basicMetricForm.value.bleu_n,
+        },
+        rouge: {
+          enabled: basicMetricForm.value.rouge_enabled,
+          methods: basicMetricForm.value.rouge_methods,
+        },
+        cosine: {
+          enabled: basicMetricForm.value.cosine_enabled,
+        },
+        output_precision: basicMetricForm.value.output_precision,
+      },
     })
     ElMessage.success('评测任务已创建并启动')
     router.push('/model-eval')
@@ -168,17 +175,33 @@ async function handleSubmit() {
 
 async function handleNext() {
   if (loading.value || submitting.value) return
-  const taskValid = await taskStepRef.value?.validate()
+  if (currentStep.value === 0) {
-  if (!taskValid) {
+    const taskValid = await taskStepRef.value?.validate()
-    ElMessage.warning('请检查并完善任务配置')
+    if (!taskValid) {
-    return
+      ElMessage.warning('请检查并完善任务配置')
+      return
+    }
   }
-  currentStep.value = 1
+  if (currentStep.value === 1) {
+    const ruleValid = await ruleStepRef.value?.validate()
+    if (!ruleValid) {
+      ElMessage.warning('请检查并完善大模型评测指标')
+      return
+    }
+  }
+  if (currentStep.value === 2) {
+    const basicMetricValid = await basicMetricStepRef.value?.validate()
+    if (!basicMetricValid) {
+      ElMessage.warning('请检查并完善基础评测指标')
+      return
+    }
+  }
+  currentStep.value = Math.min(currentStep.value + 1, WIZARD_STEPS.length - 1)
 }
 
 function handleBack() {
   if (submitting.value) return
-  currentStep.value = 0
+  currentStep.value = Math.max(currentStep.value - 1, 0)
 }
 
 function handleCancel() {
@@ -189,7 +212,10 @@ onMounted(loadData)
 </script>
 
 <template>
-  <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">
+  <PageCard
+    title="新建评测任务"
+    subtitle="依次配置任务、大模型评测指标与基础评测指标，确认后开始评测"
+  >
     <div class="create-wizard-layout">
       <main v-loading="loading" class="wizard-main">
         <div class="wizard-steps-container" aria-label="评测任务创建步骤">
@@ -207,6 +233,7 @@ onMounted(loadData)
                   'is-completed': currentStep > index,
                 }"
                 :aria-current="currentStep === index ? 'step' : undefined"
+                :aria-label="`${index + 1}. ${step.title}：${step.description}`"
               >
                 <div class="step-node">
                   <div class="step-icon" aria-hidden="true">
@@ -224,48 +251,64 @@ onMounted(loadData)
         </div>
 
         <div class="wizard-content">
-        <EvalTaskSetupStep
+          <EvalTaskSetupStep
-          v-if="currentStep === 0"
+            v-if="currentStep === 0"
-          ref="taskStepRef"
+            ref="taskStepRef"
-          v-model="taskForm"
+            v-model="taskForm"
-          :trained-models="trainedModels"
+            :trained-models="trainedModels"
-          :eval-datasets="evalDatasets"
+            :eval-datasets="evalDatasets"
-          :gpus="gpus"
+            :gpus="gpus"
-          :loading="loading"
+            :loading="loading"
-          :disabled="submitting"
+            :disabled="submitting"
-        />
+          />
-        <EvalRuleSetupStep
+          <EvalRuleSetupStep
-          v-else
+            v-else-if="currentStep === 1"
-          ref="ruleStepRef"
+            ref="ruleStepRef"
-          v-model="ruleForm"
+            v-model="ruleForm"
-          :dimensions="dimensions"
+            :eval-models="evalModels"
-          :eval-models="evalModels"
+            :disabled="submitting"
-          :loading="loading"
+          />
-          :disabled="submitting"
+          <BasicMetricSetupStep
-        />
+            v-else-if="currentStep === 2"
+            ref="basicMetricStepRef"
+            v-model="basicMetricForm"
+            :disabled="submitting"
+          />
+          <StartEvalStep
+            v-else
+            :task="taskForm"
+            :llm-metric="ruleForm"
+            :basic-metrics="basicMetricForm"
+            :trained-models="trainedModels"
+            :eval-datasets="evalDatasets"
+            :eval-models="evalModels"
+            :gpus="gpus"
+          />
         </div>
       </main>
 
       <footer class="wizard-footer">
         <el-button
-          v-if="currentStep === 1"
+          v-if="currentStep > 0"
           class="footer-back"
           :disabled="submitting"
           @click="handleBack"
         >
-          <i class="fa fa-arrow-left footer-button-icon" />返回：任务配置
+          <i class="fa fa-arrow-left footer-button-icon" />返回：{{ WIZARD_STEPS[currentStep - 1].title }}
         </el-button>
         <el-button v-else class="footer-back" :disabled="submitting" @click="handleCancel">
           取消
         </el-button>
 
         <el-button
-          v-if="currentStep === 0"
+          v-if="currentStep < WIZARD_STEPS.length - 1"
           type="primary"
           :disabled="loading || submitting"
           @click="handleNext"
         >
-          下一步：评测规则<i class="fa fa-arrow-right footer-button-icon is-right" />
+          下一步：{{ WIZARD_STEPS[currentStep + 1].title }}<i
+            class="fa fa-arrow-right footer-button-icon is-right"
+          />
         </el-button>
         <el-button
           v-else
@@ -274,10 +317,10 @@ onMounted(loadData)
           :disabled="loading || submitting"
           @click="handleSubmit"
         >
-          创建并启动评测
+          开始评测
         </el-button>
       </footer>
-      </div>
+    </div>
   </PageCard>
 </template>
 
@@ -305,7 +348,7 @@ onMounted(loadData)
   display: flex;
   align-items: center;
   justify-content: space-between;
-  max-width: 760px;
+  max-width: 1120px;
   margin: 0 auto;
   padding: 0 20px;
 }
@@ -319,7 +362,7 @@ onMounted(loadData)
 .step-connector {
   flex: 1;
   height: 2px;
-  margin: 0 20px;
+  margin: 0 12px;
   background: #e2e8f0;
   transition: background-color 0.2s ease;
 }
@@ -414,6 +457,12 @@ onMounted(loadData)
   margin-left: 6px;
 }
 
+@media (max-width: 1100px) {
+  .step-description {
+    display: none;
+  }
+}
+
 @media (max-width: 760px) {
   .wizard-main {
     padding: 24px 20px;
@@ -424,10 +473,10 @@ onMounted(loadData)
   }
 
   .step-connector {
-    margin: 0 10px;
+    margin: 0 8px;
   }
 
-  .step-description {
+  .step-text {
     display: none;
   }
 

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

@@ -21,19 +21,28 @@ export interface DimensionFormDraft {
   pass_threshold: number
 }
 
-withDefaults(
+const props = withDefaults(
   defineProps<{
     evalModels: ModelItem[]
     showDescription?: boolean
+    showStatusSettings?: boolean
+    allowedTypes?: DimensionType[]
   }>(),
   {
     showDescription: true,
+    showStatusSettings: true,
+    allowedTypes: () => ['classification', 'metric', 'text_similarity'],
   },
 )
 
 const form = defineModel<DimensionFormDraft>({ required: true })
+const SCORE_STEP = 0.5
 
-const typeOptions = Object.entries(DIMENSION_TYPE_MAP).map(([value, label]) => ({ value, label }))
+const typeOptions = computed(() =>
+  Object.entries(DIMENSION_TYPE_MAP)
+    .filter(([value]) => props.allowedTypes.includes(value as DimensionType))
+    .map(([value, label]) => ({ value, label })),
+)
 const currentMethods = computed(() => (form.value.type ? EVAL_METHODS[form.value.type] || [] : []))
 
 /**
@@ -65,6 +74,23 @@ watch(
   },
   { flush: 'sync' },
 )
+
+/** 始终保持评分区间合法，并把通过阈值收敛到当前区间内。 */
+watch(
+  () => [form.value.type, form.value.score_min, form.value.score_max] as const,
+  ([type, scoreMin, scoreMax]) => {
+    if (type !== 'metric') return
+    if (scoreMax <= scoreMin) {
+      form.value.score_max = scoreMin + SCORE_STEP
+      return
+    }
+    form.value.pass_threshold = Math.min(
+      Math.max(form.value.pass_threshold, scoreMin),
+      scoreMax,
+    )
+  },
+  { flush: 'sync' },
+)
 </script>
 
 <template>
@@ -134,13 +160,24 @@ watch(
   </template>
 
   <template v-if="form.type === 'metric'">
-    <el-form-item label="评分最小值">
+    <el-form-item label="评分最小值" prop="score_min">
-      <el-input-number v-model="form.score_min" :min="0" :step="1" style="width: 160px" />
+      <el-input-number
+        v-model="form.score_min"
+        :min="0"
+        :max="Math.max(form.score_max - SCORE_STEP, 0)"
+        :step="SCORE_STEP"
+        style="width: 160px"
+      />
     </el-form-item>
-    <el-form-item label="评分最大值">
+    <el-form-item label="评分最大值" prop="score_max">
-      <el-input-number v-model="form.score_max" :min="0" :step="1" style="width: 160px" />
+      <el-input-number
+        v-model="form.score_max"
+        :min="form.score_min + SCORE_STEP"
+        :step="SCORE_STEP"
+        style="width: 160px"
+      />
     </el-form-item>
-    <el-form-item label="通过阈值">
+    <el-form-item label="通过阈值" prop="pass_threshold">
       <div style="width: 100%">
         <el-slider
           v-model="form.pass_threshold"
@@ -154,12 +191,14 @@ watch(
     </el-form-item>
   </template>
 
-  <el-form-item label="启用该维度">
+  <template v-if="showStatusSettings">
-    <el-switch v-model="form.is_active" />
+    <el-form-item label="启用该维度">
-  </el-form-item>
+      <el-switch v-model="form.is_active" />
-  <el-form-item label="设为默认">
+    </el-form-item>
-    <el-switch v-model="form.is_default" />
+    <el-form-item label="设为默认">
-  </el-form-item>
+      <el-switch v-model="form.is_default" />
+    </el-form-item>
+  </template>
 </template>
 
 <style scoped>

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

@@ -1,49 +1,31 @@
 <script setup lang="ts">
-import { computed, ref } from 'vue'
+import { ref } from 'vue'
 import type { FormInstance, FormRules } from 'element-plus'
 import DimensionFormFields, { type DimensionFormDraft } from './DimensionFormFields.vue'
-import type { Dimension, ModelItem } from '@/types'
+import type { DimensionType, ModelItem } from '@/types'
 
-export type EvalRuleMode = 'existing' | 'new' | 'baseline'
+export type EvalRuleSetupDraft = DimensionFormDraft
 
-export interface EvalRuleSetupDraft {
+defineProps<{
-  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 LLM_METRIC_TYPES: DimensionType[] = ['classification', 'metric']
-const existingRules: FormRules = {
-  dimension_id: [{ required: true, message: '请选择评测维度', trigger: 'change' }],
-}
 
 const dimensionRules: FormRules<DimensionFormDraft> = {
-  type: [{ required: true, message: '请选择指标类型', trigger: 'change' }],
+  type: [{ required: true, message: '请选择大模型评测指标类型', trigger: 'change' }],
-  eval_model: [{ required: true, message: '请选择大模型', trigger: 'change' }],
+  eval_model: [{ required: true, message: '请选择评测大模型', 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() {
-  const target = newFormRef.value
+  if (!newFormRef.value) return false
-  if (!target) return false
   try {
-    const valid = await target.validate()
+    return !!(await newFormRef.value.validate())
-    return !!valid
   } catch {
     return false
   }
@@ -56,7 +38,7 @@ defineExpose({ validate })
   <div class="rule-step">
     <el-form
       ref="newFormRef"
-      :model="form.newDimension"
+      :model="form"
       :rules="dimensionRules"
       label-width="130px"
       class="step-form step-form-wide"
@@ -64,17 +46,17 @@ defineExpose({ validate })
       scroll-to-error
     >
       <DimensionFormFields
-        v-model="form.newDimension"
+        v-model="form"
         :eval-models="evalModels"
         :show-description="false"
+        :show-status-settings="false"
+        :allowed-types="LLM_METRIC_TYPES"
       />
     </el-form>
   </div>
 </template>
 
 <style scoped>
-
-
 .step-form {
   max-width: 760px;
 }
@@ -82,6 +64,4 @@ defineExpose({ validate })
 .step-form-wide {
   max-width: 920px;
 }
-
-
 </style>