feat: 评测创建支持基础指标配置与启动参数类型化

新增 StartEvalPayload 与 BasicEvalMetricsConfig 类型约束启动评测接口，创建向导补充 BLEU、ROUGE、余弦相似度等基础指标配置，维度表单与规则步骤同步调整，回归脚本扩充断言。

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',
-  dimension_id: '',
-  newDimension: {
-    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,
-  },
+  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 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[2].status === 'fulfilled') {
-    evalDatasets.value = (results[2].value || []).filter((dataset) => dataset.type === 'eval')
+  if (results[1].status === 'fulfilled') {
+    evalDatasets.value = (results[1].value || []).filter((dataset) => dataset.type === 'eval')
   }
-  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')
+  if (results[2].status === 'fulfilled') gpus.value = results[2].value?.gpu || []
+  if (results[3].status === 'fulfilled') {
+    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 (!taskValid) {
-    ElMessage.warning('请检查并完善任务配置')
-    return
+  if (currentStep.value === 0) {
+    const taskValid = await taskStepRef.value?.validate()
+    if (!taskValid) {
+      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
-          v-if="currentStep === 0"
-          ref="taskStepRef"
-          v-model="taskForm"
-          :trained-models="trainedModels"
-          :eval-datasets="evalDatasets"
-          :gpus="gpus"
-          :loading="loading"
-          :disabled="submitting"
-        />
-        <EvalRuleSetupStep
-          v-else
-          ref="ruleStepRef"
-          v-model="ruleForm"
-          :dimensions="dimensions"
-          :eval-models="evalModels"
-          :loading="loading"
-          :disabled="submitting"
-        />
+          <EvalTaskSetupStep
+            v-if="currentStep === 0"
+            ref="taskStepRef"
+            v-model="taskForm"
+            :trained-models="trainedModels"
+            :eval-datasets="evalDatasets"
+            :gpus="gpus"
+            :loading="loading"
+            :disabled="submitting"
+          />
+          <EvalRuleSetupStep
+            v-else-if="currentStep === 1"
+            ref="ruleStepRef"
+            v-model="ruleForm"
+            :eval-models="evalModels"
+            :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;
   }