feat: 数据处理向导新增生成控制选项

提取 GenerationOptionsPanel 组件统一管理生成模型、温度、最大长度、JSON 模式与质量过滤配置，StructuredProcessOptions 与 UnstructuredProcessOptions 继承 GenerationControlOptions，CreateView、TaskSetupStep 及各子步骤同步接入，详情/列表小幅调整，回归脚本扩充生成选项断言。

frontend/scripts/regression-data-process-wizard.mjs

@@ -126,6 +126,8 @@ assert.match(previewSource, /@media \(max-width: 900px\)/, '第三步缺少窄
 
 const taskSetupPath = path.join(createDir, 'TaskSetupStep.vue')
 const taskSetupSource = await readFile(taskSetupPath, 'utf8')
+const generationOptionsPath = path.join(createDir, 'GenerationOptionsPanel.vue')
+const generationControlSource = await readFile(generationOptionsPath, 'utf8')
 const sourceUploadPath = path.join(createDir, 'SourceUploadStep.vue')
 const sourceUploadSource = await readFile(sourceUploadPath, 'utf8')
 
@@ -135,7 +137,7 @@ for (const marker of ['<el-upload', '源数据上传', '数据源配置', 'uploa
 assert.match(viewSource, /<SourceUploadStep\s+[\s\S]*?v-else-if="currentStepId === 'upload'"/, '第二步没有挂载独立上传组件')
 assert.match(viewSource, /if \(currentStepId\.value === 'create'\) return '继续：上传文件'/, '第一步主按钮没有指向上传文件')
 assert.match(viewSource, /if \(currentStepId\.value === 'upload'\) return '继续：数据预览'/, '第二步主按钮没有指向数据预览')
-assert.match(viewSource, /const DRAFT_SCHEMA_VERSION = 4/, '五步索引变更后必须升级草稿版本')
+assert.match(viewSource, /const DRAFT_SCHEMA_VERSION = 5/, '生成配置扩展后必须升级草稿版本')
 
 const nextFromCreateStart = viewSource.indexOf('async function nextFromCreate()')
 const nextFromUploadStart = viewSource.indexOf('function nextFromUpload()', nextFromCreateStart)
@@ -195,6 +197,40 @@ assert.match(viewSource, /structuredOptions\.value = \{[\s\S]*\.\.\.snapshot\.st
 assert.match(viewSource, /v-model:structured-options="structuredOptions"/, '父页面没有双向绑定结构化配置')
 assert.match(viewSource, /createResults\([\s\S]*structuredOptions\.value/, '每行生成数量没有接入结果生成逻辑')
 
+for (const field of [
+  'generationModelId',
+  'generationPrompt',
+  'qualityFilterEnabled',
+  'filterLowQuality',
+  'filterShortContent',
+  'minOutputLength',
+]) {
+  assert.ok(typesSource.includes(field), `生成控制配置缺少字段：${field}`)
+  assert.ok(viewSource.includes(field), `父页面默认值或草稿状态缺少字段：${field}`)
+}
+assert.match(taskSetupSource, /GenerationOptionsPanel/, '生成选项没有复用统一的大模型与质量筛选组件')
+assert.match(taskSetupSource, /:options="structuredOptions"/, '结构化生成选项未接入统一配置组件')
+assert.match(taskSetupSource, /:options="unstructuredOptions"/, '非结构化生成选项未接入统一配置组件')
+assert.match(taskSetupSource, /<h3>大模型<\/h3>/, '大模型必须作为与生成选项平级的独立分类')
+assert.match(taskSetupSource, /section="model"/, '独立大模型分类没有挂载模型配置')
+assert.match(taskSetupSource, /section="quality"/, '质量筛选没有保留在生成选项分类中')
+assert.doesNotMatch(generationControlSource, /<h4>大模型<\/h4>/, '大模型不应继续作为生成选项内部子分类')
+for (const label of ['大模型', '数据生成模型', '默认提示语', '质量筛选', '过滤低质量内容', '过滤过短内容', '最少字数']) {
+  assert.ok(generationControlSource.includes(label), `生成控制界面缺少：${label}`)
+}
+assert.match(generationControlSource, /filterable/, '数据生成模型下拉必须支持搜索')
+assert.match(generationControlSource, /maxlength="500"/, '默认提示语缺少合理的长度限制')
+assert.match(generationControlSource, /\.model-option-row\s*\{[\s\S]*?display:\s*flex[\s\S]*?border:\s*1px solid #e2e5ec/, '大模型配置没有与上方生成选项使用一致的配置行样式')
+assert.match(viewSource, /const DEFAULT_GENERATION_PROMPT\s*=\s*['"][^'"]{40,}['"]/, '大模型配置缺少可直接使用的默认提示语')
+assert.equal((viewSource.match(/generationPrompt:\s*DEFAULT_GENERATION_PROMPT/g) || []).length, 2, '结构化与非结构化任务必须共用默认提示语')
+assert.match(generationControlSource, /v-if="options\.qualityFilterEnabled"/, '质量规则没有随总开关渐进显示')
+assert.match(generationControlSource, /v-if="options\.filterShortContent"/, '最少字数没有随短内容规则显示')
+assert.match(generationControlSource, /:min="1"[\s\S]*:max="1000"/, '最少字数缺少 1 到 1000 的边界限制')
+assert.match(taskSetupSource, /generationValidationMessage/, '生成模型与质量规则缺少继续前校验')
+assert.match(viewSource, /useModelsStore/, '创建页没有加载模型列表')
+assert.match(viewSource, /model\.type === 'LLM'/, '数据生成模型列表没有排除非大模型')
+assert.match(viewSource, /:generation-models="generationModels"/, '创建页没有向生成选项传递模型列表')
+
 assert.match(typesSource, /export interface UnstructuredProcessOptions/, '缺少非结构化处理选项类型')
 for (const field of [
   'preprocessOptions',
@@ -298,7 +334,17 @@ const generationOptionsStart = viewSource.indexOf('function generationAffectingO
 const generationOptionsEnd = viewSource.indexOf('const generationOptionsSignature', generationOptionsStart)
 assert.ok(generationOptionsStart >= 0 && generationOptionsEnd > generationOptionsStart, '缺少生成影响配置签名函数')
 const generationOptionsSource = viewSource.slice(generationOptionsStart, generationOptionsEnd)
-for (const field of ['semanticEnrichment', 'qaPairsPerChunk', 'datasetSplit']) {
+for (const field of [
+  'semanticEnrichment',
+  'qaPairsPerChunk',
+  'datasetSplit',
+  'generationModelId',
+  'generationPrompt',
+  'qualityFilterEnabled',
+  'filterLowQuality',
+  'filterShortContent',
+  'minOutputLength',
+]) {
   assert.ok(generationOptionsSource.includes(field), `生成签名缺少字段：${field}`)
 }
 assert.match(
@@ -341,6 +387,12 @@ const baseUnstructuredOptions = {
   semanticEnrichment: false,
   qaPairsPerChunk: 3,
   datasetSplit: { train: 80, validation: 10, test: 10 },
+  generationModelId: 1,
+  generationPrompt: '仅输出问答对',
+  qualityFilterEnabled: false,
+  filterLowQuality: true,
+  filterShortContent: true,
+  minOutputLength: 20,
 }
 
 for (const chunkMethod of ['semantic', 'heading', 'fixed', 'custom']) {
@@ -434,6 +486,31 @@ assert.ok(samplePreviewItems.length > 12, '测试文档未生成足够的切片'
 const generatedResults = previewModelModule.createResults(samplePreviewItems.slice(0, 13), baseUnstructuredOptions)
 assert.equal(generatedResults.length, 39, '每个切片生成 3 个问答对未完整应用到所有切片')
 
+const shortContentItems = [{
+  ...samplePreviewItems[0],
+  editedContent: '问：示例\n短回答',
+}]
+const filteredShortResults = previewModelModule.createResults(shortContentItems, {
+  ...baseUnstructuredOptions,
+  qualityFilterEnabled: true,
+  filterLowQuality: false,
+  filterShortContent: true,
+  minOutputLength: 20,
+})
+assert.equal(filteredShortResults.length, 0, '开启过短内容过滤后仍保留低于最少字数的结果')
+
+const invalidContentItems = [{
+  ...samplePreviewItems[0],
+  status: 'invalid',
+}]
+const filteredInvalidResults = previewModelModule.createResults(invalidContentItems, {
+  ...baseUnstructuredOptions,
+  qualityFilterEnabled: true,
+  filterLowQuality: true,
+  filterShortContent: false,
+})
+assert.equal(filteredInvalidResults.length, 0, '开启低质量过滤后仍保留标记为无效的结果')
+
 const legacyExternalItems = previewModelModule.buildPreviewItems('a\nb\nc\nd', 'external', 'legacy-check')
 assert.equal(legacyExternalItems.length, 2, '外来数据原有的每 3 行分组行为被破坏')
 

frontend/src/views/data-process/DataProcessCreateView.vue

@@ -2,6 +2,7 @@
 import { computed, nextTick, onBeforeUnmount, onMounted, reactive, ref, watch } from 'vue'
 import { onBeforeRouteLeave, useRouter } from 'vue-router'
 import { ElMessage, type UploadFile } from 'element-plus'
+import { storeToRefs } from 'pinia'
 import AppConfirmDialog from '@/components/AppConfirmDialog.vue'
 import TaskSetupStep from './create/TaskSetupStep.vue'
 import SourceUploadStep from './create/SourceUploadStep.vue'
@@ -9,6 +10,7 @@ import PreviewCompareStep from './create/PreviewCompareStep.vue'
 import GenerationStep from './create/GenerationStep.vue'
 import ResultEditorStep from './create/ResultEditorStep.vue'
 import { buildPreviewItems, createResults, DEFAULT_SOURCE_TEXT } from './create/previewModel'
+import { useModelsStore } from '@/stores/models'
 import type {
   ExternalDataSource,
   GenerationState,
@@ -21,11 +23,15 @@ import type {
 } from './create/types'
 
 const router = useRouter()
+const modelsStore = useModelsStore()
+const { list: modelList } = storeToRefs(modelsStore)
+const generationModels = computed(() => modelList.value.filter((model) => model.type === 'LLM'))
 const taskSetupRef = ref<InstanceType<typeof TaskSetupStep>>()
 const confirmDialogRef = ref<InstanceType<typeof AppConfirmDialog>>()
 const DRAFT_STORAGE_KEY = 'yg-data-process-create-draft'
-const DRAFT_SCHEMA_VERSION = 4
+const DRAFT_SCHEMA_VERSION = 5
 const PREVIEW_MODEL_VERSION = 'document-chunk-v2'
+const DEFAULT_GENERATION_PROMPT = '你是一名专业的数据生成助手。请根据输入内容生成准确、完整、可直接用于模型训练的问答数据。仅输出符合目标格式的内容，答案应事实清晰、语言自然，不要添加分析过程、说明或无关内容。'
 
 const WIZARD_STEPS = [
   { id: 'create', title: '创建任务', desc: '填写任务信息与处理配置' },
@@ -45,6 +51,15 @@ const structuredOptions = ref<StructuredProcessOptions>({
   semanticEnrichment: false,
   qaPairsPerRow: 1,
   datasetSplit: { train: 80, validation: 10, test: 10 },
+  generationModelId: '',
+  generationPrompt: DEFAULT_GENERATION_PROMPT,
+  temperature: 0.7,
+  maxTokens: 1024,
+  jsonMode: false,
+  qualityFilterEnabled: false,
+  filterLowQuality: true,
+  filterShortContent: true,
+  minOutputLength: 20,
 })
 const unstructuredOptions = ref<UnstructuredProcessOptions>({
   preprocessOptions: [
@@ -66,6 +81,15 @@ const unstructuredOptions = ref<UnstructuredProcessOptions>({
   semanticEnrichment: false,
   qaPairsPerChunk: 1,
   datasetSplit: { train: 80, validation: 10, test: 10 },
+  generationModelId: '',
+  generationPrompt: DEFAULT_GENERATION_PROMPT,
+  temperature: 0.7,
+  maxTokens: 1024,
+  jsonMode: false,
+  qualityFilterEnabled: false,
+  filterLowQuality: true,
+  filterShortContent: true,
+  minOutputLength: 20,
 })
 interface UploadedDataFile {
   uid: number | string
@@ -219,19 +243,45 @@ function previewAffectingOptions() {
 
 function generationAffectingOptions() {
   if (processType.value === 'structured') {
-    const { semanticEnrichment, qaPairsPerRow, datasetSplit } = structuredOptions.value
+    const {
-    return { semanticEnrichment, qaPairsPerRow, datasetSplit }
+      semanticEnrichment, qaPairsPerRow, datasetSplit, generationModelId, generationPrompt,
+      temperature, maxTokens, jsonMode,
+      qualityFilterEnabled, filterLowQuality, filterShortContent, minOutputLength,
+    } = structuredOptions.value
+    return {
+      semanticEnrichment, qaPairsPerRow, datasetSplit, generationModelId, generationPrompt,
+      temperature, maxTokens, jsonMode,
+      qualityFilterEnabled, filterLowQuality, filterShortContent, minOutputLength,
+    }
   }
   if (processType.value === 'unstructured') {
     const {
       semanticEnrichment,
       qaPairsPerChunk,
       datasetSplit,
+      generationModelId,
+      generationPrompt,
+      temperature,
+      maxTokens,
+      jsonMode,
+      qualityFilterEnabled,
+      filterLowQuality,
+      filterShortContent,
+      minOutputLength,
     } = unstructuredOptions.value
     return {
       semanticEnrichment,
       qaPairsPerChunk,
       datasetSplit,
+      generationModelId,
+      generationPrompt,
+      temperature,
+      maxTokens,
+      jsonMode,
+      qualityFilterEnabled,
+      filterLowQuality,
+      filterShortContent,
+      minOutputLength,
     }
   }
   return null
@@ -337,6 +387,31 @@ function restoreDraft() {
         qaPairsPerChunk: Number.isFinite(restoredUnstructuredOptions.qaPairsPerChunk)
           ? restoredUnstructuredOptions.qaPairsPerChunk
           : unstructuredOptions.value.qaPairsPerChunk,
+        generationModelId: restoredUnstructuredOptions.generationModelId ?? unstructuredOptions.value.generationModelId,
+        generationPrompt: typeof restoredUnstructuredOptions.generationPrompt === 'string'
+          ? restoredUnstructuredOptions.generationPrompt
+          : unstructuredOptions.value.generationPrompt,
+        temperature: Number.isFinite(restoredUnstructuredOptions.temperature)
+          ? restoredUnstructuredOptions.temperature
+          : unstructuredOptions.value.temperature,
+        maxTokens: Number.isFinite(restoredUnstructuredOptions.maxTokens)
+          ? restoredUnstructuredOptions.maxTokens
+          : unstructuredOptions.value.maxTokens,
+        jsonMode: typeof restoredUnstructuredOptions.jsonMode === 'boolean'
+          ? restoredUnstructuredOptions.jsonMode
+          : unstructuredOptions.value.jsonMode,
+        qualityFilterEnabled: typeof restoredUnstructuredOptions.qualityFilterEnabled === 'boolean'
+          ? restoredUnstructuredOptions.qualityFilterEnabled
+          : unstructuredOptions.value.qualityFilterEnabled,
+        filterLowQuality: typeof restoredUnstructuredOptions.filterLowQuality === 'boolean'
+          ? restoredUnstructuredOptions.filterLowQuality
+          : unstructuredOptions.value.filterLowQuality,
+        filterShortContent: typeof restoredUnstructuredOptions.filterShortContent === 'boolean'
+          ? restoredUnstructuredOptions.filterShortContent
+          : unstructuredOptions.value.filterShortContent,
+        minOutputLength: Number.isFinite(restoredUnstructuredOptions.minOutputLength)
+          ? restoredUnstructuredOptions.minOutputLength
+          : unstructuredOptions.value.minOutputLength,
         datasetSplit: {
           train: Number.isFinite(restoredDatasetSplit?.train)
             ? restoredDatasetSplit.train
@@ -819,7 +894,10 @@ onBeforeRouteLeave(async () => {
 })
 
 onBeforeUnmount(stopGenerationTimer)
-onMounted(restoreDraft)
+onMounted(() => {
+  restoreDraft()
+  modelsStore.load()
+})
 </script>
 
 <template>
@@ -861,6 +939,7 @@ onMounted(restoreDraft)
             v-model:process-type="processType"
             v-model:structured-options="structuredOptions"
             v-model:unstructured-options="unstructuredOptions"
+            :generation-models="generationModels"
           />
 
           <SourceUploadStep

frontend/src/views/data-process/DataProcessDetailView.vue

@@ -18,7 +18,6 @@ interface ProcessDetail {
   status: string
   description: string
   processType: string
-  dataType: string
   sourceDataset: string
   outputDataset?: string
   outputDatasetId?: number
@@ -75,8 +74,7 @@ const detailMap: Record<string, ProcessDetail> = {
     name: '客服问答数据清洗',
     status: 'completed',
     description: '清洗客服对话中的无效记录、重复问答和格式异常内容，生成可用于模型训练的标准数据集。',
-    processType: '去重清洗',
+    processType: '结构化数据',
-    dataType: '结构化数据',
     sourceDataset: '客服对话原始集',
     outputDataset: '客服对话清洗集',
     outputDatasetId: 7,
@@ -101,7 +99,7 @@ const detailMap: Record<string, ProcessDetail> = {
   },
   '492015': {
     id: '492015', name: '指令微调数据构造', status: 'running',
-    description: '从通用语料中构造指令微调训练样本。', processType: '指令构造', dataType: '非结构化数据',
+    description: '从通用语料中构造指令微调训练样本。', processType: '非结构化数据',
     sourceDataset: '通用语料库', outputDataset: 'SFT 指令集', outputDatasetId: 8, creator: '管理员',
     createTime: '2026-07-09 09:10:00', startTime: '2026-07-09 09:10:21', duration: '处理中', progress: 68,
     inputCount: 18500, outputCount: 8568, filteredCount: 425, duplicateCount: 192, errorCount: 8,
@@ -112,7 +110,7 @@ const detailMap: Record<string, ProcessDetail> = {
   },
   '731948': {
     id: '731948', name: '敏感信息脱敏处理', status: 'pending', description: '识别并脱敏用户反馈数据中的敏感字段。',
-    processType: '脱敏处理', dataType: '结构化数据', sourceDataset: '用户反馈数据', outputDataset: '用户反馈脱敏集',
+    processType: '结构化数据', sourceDataset: '用户反馈数据', outputDataset: '用户反馈脱敏集',
     outputDatasetId: 9, creator: '管理员', createTime: '2026-07-09 16:45:00', duration: '等待执行', progress: 0,
     inputCount: 9340, outputCount: 0, filteredCount: 0, duplicateCount: 0, errorCount: 0,
     config: [
@@ -122,7 +120,7 @@ const detailMap: Record<string, ProcessDetail> = {
   },
   '582012': {
     id: '582012', name: '多轮对话拼接', status: 'failed', description: '将单轮问答按会话标识拼接为多轮对话数据。',
-    processType: '格式转换', dataType: '结构化数据', sourceDataset: '单轮问答集', creator: '管理员',
+    processType: '结构化数据', sourceDataset: '单轮问答集', creator: '管理员',
     createTime: '2026-07-10 08:30:00', startTime: '2026-07-10 08:30:16', completeTime: '2026-07-10 08:31:04',
     duration: '48 秒', progress: 37, inputCount: 7520, outputCount: 2780, filteredCount: 24, duplicateCount: 0, errorCount: 1,
     config: [
@@ -186,7 +184,6 @@ function resetPage() {
         <dl class="heading-meta">
           <div><dt>任务 ID</dt><dd>{{ detail.id }}</dd></div>
           <div><dt>处理类型</dt><dd>{{ detail.processType }}</dd></div>
-          <div><dt>数据类型</dt><dd>{{ detail.dataType }}</dd></div>
           <div><dt>创建人</dt><dd>{{ detail.creator }}</dd></div>
         </dl>
       </div>

frontend/src/views/data-process/DataProcessListView.vue

@@ -5,17 +5,25 @@ import { ElMessage } from 'element-plus'
 import DataTablePage from '@/components/DataTablePage.vue'
 import ModelStatusTag from '@/components/ModelStatusTag.vue'
 
+import type { ProcessType } from './create/types'
+
 /** 数据处理任务类型 */
 interface DataProcessTask {
   id: number | string
   name: string
   status: string
-  process_type: string
+  process_type: ProcessType
   source_dataset: string
   output_dataset?: string
   create_time?: string
 }
 
+const processTypeMap: Record<ProcessType, string> = {
+  structured: '结构化数据',
+  unstructured: '非结构化数据',
+  external: '外来数据源拉取',
+}
+
 // TODO: 接入真实接口前，先用本地 mock 数据
 const router = useRouter()
 const dataList = ref<DataProcessTask[]>([
@@ -23,7 +31,7 @@ const dataList = ref<DataProcessTask[]>([
     id: 183921,
     name: '客服问答数据清洗',
     status: 'completed',
-    process_type: '去重清洗',
+    process_type: 'structured',
     source_dataset: '客服对话原始集',
     output_dataset: '客服对话清洗集',
     create_time: '2026-07-08 14:23:00',
@@ -32,7 +40,7 @@ const dataList = ref<DataProcessTask[]>([
     id: 492015,
     name: '指令微调数据构造',
     status: 'running',
-    process_type: '指令构造',
+    process_type: 'unstructured',
     source_dataset: '通用语料库',
     output_dataset: 'SFT 指令集',
     create_time: '2026-07-09 09:10:00',
@@ -41,7 +49,7 @@ const dataList = ref<DataProcessTask[]>([
     id: 731948,
     name: '敏感信息脱敏处理',
     status: 'pending',
-    process_type: '脱敏处理',
+    process_type: 'structured',
     source_dataset: '用户反馈数据',
     create_time: '2026-07-09 16:45:00',
   },
@@ -49,7 +57,7 @@ const dataList = ref<DataProcessTask[]>([
     id: 582012,
     name: '多轮对话拼接',
     status: 'failed',
-    process_type: '格式转换',
+    process_type: 'structured',
     source_dataset: '单轮问答集',
     create_time: '2026-07-10 08:30:00',
   },
@@ -98,7 +106,12 @@ function formatDateTime(value?: string) {
           </template>
         </el-table-column>
         <el-table-column label="处理类型" align="center" width="140">
-          <template #default="{ row }">{{ row.process_type || '-' }}</template>
+          <template #default="{ row }">
+            <el-tag v-if="row.process_type" size="small" type="info" effect="plain">
+              {{ processTypeMap[row.process_type as ProcessType] || row.process_type }}
+            </el-tag>
+            <span v-else>-</span>
+          </template>
         </el-table-column>
         <el-table-column label="源数据集" align="center" show-overflow-tooltip>
           <template #default="{ row }">{{ row.source_dataset || '-' }}</template>

frontend/src/views/data-process/create/GenerationOptionsPanel.vue

@@ -0,0 +1,439 @@
+<script setup lang="ts">
+import { ref } from 'vue'
+import type { ModelItem } from '@/types'
+import type { GenerationControlOptions } from './types'
+
+const props = defineProps<{
+  options: GenerationControlOptions
+  models: ModelItem[]
+  section: 'model' | 'quality'
+  validationMessage?: string
+}>()
+
+const advancedModelSettingsOpen = ref(false)
+
+const emit = defineEmits<{
+  'update:options': [value: GenerationControlOptions]
+}>()
+
+function updateField<K extends keyof GenerationControlOptions>(
+  field: K,
+  value: GenerationControlOptions[K],
+) {
+  emit('update:options', { ...props.options, [field]: value })
+}
+
+function updateQualityRules(value: Array<string | number>) {
+  const rules = value.filter((item): item is string => typeof item === 'string')
+  emit('update:options', {
+    ...props.options,
+    filterLowQuality: rules.includes('low_quality'),
+    filterShortContent: rules.includes('short_content'),
+  })
+}
+
+const selectedQualityRules = () => [
+  props.options.filterLowQuality ? 'low_quality' : '',
+  props.options.filterShortContent ? 'short_content' : '',
+].filter(Boolean)
+
+function sectionValidationMessage() {
+  if (!props.validationMessage) return ''
+  const isModelMessage = props.validationMessage.includes('数据生成模型')
+  if (props.section === 'model') return isModelMessage ? props.validationMessage : ''
+  return isModelMessage ? '' : props.validationMessage
+}
+</script>
+
+<template>
+  <div v-if="section === 'model'" class="generation-config-group model-config-group">
+    <div class="model-field">
+      <div class="field-copy">
+        <strong>数据生成模型 <em>*</em></strong>
+        <small>选择负责生成问答、指令或结构化内容的大模型</small>
+      </div>
+      <el-select
+        class="model-select"
+        :model-value="options.generationModelId"
+        filterable
+        placeholder="请选择用于生成数据的模型"
+        aria-label="数据生成模型"
+        @update:model-value="updateField('generationModelId', $event)"
+      >
+        <el-option
+          v-for="model in models"
+          :key="model.id"
+          :label="model.name"
+          :value="model.id"
+        >
+          <div class="model-option">
+            <span>{{ model.name }}</span>
+            <small>{{ model.model_source === 'api' ? '在线模型' : '本地模型' }}</small>
+          </div>
+        </el-option>
+        <template #empty>
+          <div class="model-empty">暂无可用的大模型，请先在模型管理中添加</div>
+        </template>
+      </el-select>
+    </div>
+
+    <div class="model-field">
+      <div class="field-copy">
+        <strong>默认提示语</strong>
+        <small>用于约束生成内容的格式、语气和完整性，可按任务需要修改</small>
+      </div>
+      <el-input
+        class="prompt-input"
+        :model-value="options.generationPrompt"
+        type="textarea"
+        :rows="4"
+        maxlength="500"
+        show-word-limit
+        resize="vertical"
+        placeholder="请输入模型生成内容时需要遵循的要求"
+        aria-label="模型默认提示语"
+        @update:model-value="updateField('generationPrompt', $event)"
+      />
+      <div class="prompt-variables-hint">
+        提示：可在文本中通过 <code>{{ content }}</code> 引用当前正在处理的数据内容。
+      </div>
+    </div>
+
+    <button
+      type="button"
+      class="advanced-settings-toggle"
+      :aria-expanded="advancedModelSettingsOpen"
+      @click="advancedModelSettingsOpen = !advancedModelSettingsOpen"
+    >
+      <span>
+        <strong>高级设置</strong>
+        <small>温度、最大输出长度与格式约束</small>
+      </span>
+      <i class="fa" :class="advancedModelSettingsOpen ? 'fa-chevron-up' : 'fa-chevron-down'" />
+    </button>
+
+    <div v-if="advancedModelSettingsOpen" class="advanced-settings-panel">
+      <div class="advanced-settings-grid">
+        <label class="config-field">
+          <span class="config-field-label">生成温度 (Temperature)</span>
+          <el-slider
+            :model-value="options.temperature"
+            :min="0"
+            :max="1"
+            :step="0.1"
+            show-input
+            @update:model-value="updateField('temperature', Number($event) || 0)"
+          />
+          <small>较低值输出更确定、保守，较高值输出更多样、有创造性</small>
+        </label>
+
+        <label class="config-field">
+          <span class="config-field-label">最大输出长度 (Max Tokens)</span>
+          <span class="unit-input">
+            <el-input-number
+              :model-value="options.maxTokens"
+              :min="100"
+              :max="8192"
+              :step="256"
+              controls-position="right"
+              @update:model-value="updateField('maxTokens', Number($event) || 1024)"
+            />
+            <span>Token</span>
+          </span>
+          <small>限制单次生成的最大内容长度，防止过度发散</small>
+        </label>
+      </div>
+
+      <div class="json-mode-row">
+        <div class="field-copy">
+          <strong>强制 JSON 格式输出</strong>
+          <small>要求大模型返回纯 JSON 对象，有助于提高下游结构化提取的稳定性</small>
+        </div>
+        <el-switch
+          :model-value="options.jsonMode"
+          inline-prompt
+          active-text="开"
+          inactive-text="关"
+          @update:model-value="updateField('jsonMode', Boolean($event))"
+        />
+      </div>
+    </div>
+
+    <p v-if="sectionValidationMessage()" class="generation-validation-message" role="alert">
+      {{ sectionValidationMessage() }}
+    </p>
+  </div>
+
+  <div v-else class="generation-config-group quality-config-group">
+    <div class="quality-switch-row">
+      <div class="config-group-heading">
+        <div>
+          <h4>质量筛选</h4>
+          <p>生成后自动过滤不符合要求的内容，不影响源数据预处理</p>
+        </div>
+      </div>
+      <el-switch
+        :model-value="options.qualityFilterEnabled"
+        inline-prompt
+        active-text="开"
+        inactive-text="关"
+        aria-label="质量筛选"
+        @update:model-value="updateField('qualityFilterEnabled', Boolean($event))"
+      />
+    </div>
+
+    <div v-if="options.qualityFilterEnabled" class="quality-rule-panel">
+      <el-checkbox-group
+        :model-value="selectedQualityRules()"
+        class="quality-rule-list"
+        @update:model-value="updateQualityRules"
+      >
+        <el-checkbox value="low_quality">
+          <span class="rule-copy">
+            <strong>过滤低质量内容</strong>
+            <small>按相关性、完整性和可读性过滤明显不合格的生成结果</small>
+          </span>
+        </el-checkbox>
+        <el-checkbox value="short_content">
+          <span class="rule-copy">
+            <strong>过滤过短内容</strong>
+            <small>过滤正文长度低于指定字数的生成结果</small>
+          </span>
+        </el-checkbox>
+      </el-checkbox-group>
+
+      <label v-if="options.filterShortContent" class="minimum-length-field">
+        <span>最少字数</span>
+        <span class="minimum-length-input">
+          <el-input-number
+            :model-value="options.minOutputLength"
+            :min="1"
+            :max="1000"
+            :step="5"
+            :precision="0"
+            controls-position="right"
+            aria-label="生成内容最少字数"
+            @update:model-value="updateField('minOutputLength', Number($event) || 1)"
+          />
+          <span>字</span>
+        </span>
+        <small>少于该字数的生成结果将被过滤</small>
+      </label>
+
+      <p v-if="!options.filterLowQuality && !options.filterShortContent" class="quality-rule-warning" role="alert">
+        请至少选择一项质量筛选规则
+      </p>
+    </div>
+    <p v-if="sectionValidationMessage()" class="generation-validation-message" role="alert">
+      {{ sectionValidationMessage() }}
+    </p>
+  </div>
+</template>
+
+<style scoped lang="scss">
+.generation-config-group {
+  padding: 14px;
+  border: 1px solid #e2e5ec;
+  border-radius: 8px;
+  background: #fff;
+}
+
+.config-group-heading {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 16px;
+
+  h4 { margin: 0; color: #344054; font-size: 13px; font-weight: 600; }
+  p { margin: 4px 0 0; color: #8a93a3; font-size: 12px; line-height: 1.5; }
+}
+
+.model-config-group {
+  display: flex;
+  flex-direction: column;
+  gap: 16px;
+}
+
+.model-field {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+}
+
+.field-copy {
+  display: grid;
+  gap: 4px;
+
+  strong {
+    color: #344054;
+    font-size: 13px;
+    font-weight: 600;
+  }
+
+  small {
+    color: #8a93a3;
+    font-size: 12px;
+    line-height: 1.5;
+  }
+
+  em { color: var(--el-color-danger); font-style: normal; }
+}
+
+.model-select { width: 100%; }
+.prompt-input { width: 100%; }
+.model-option { width: 100%; display: flex; align-items: center; justify-content: space-between; gap: 16px; }
+.model-option small { color: #98a2b3; font-size: 11px; }
+.model-empty { padding: 12px; color: #8a93a3; font-size: 12px; text-align: center; }
+
+.prompt-variables-hint {
+  color: #8a93a3;
+  font-size: 12px;
+  code {
+    padding: 2px 4px;
+    color: #5b50f2;
+    background: #f0f0ff;
+    border-radius: 4px;
+    font-family: inherit;
+  }
+}
+
+.advanced-settings-toggle {
+  display: flex;
+  width: 100%;
+  min-height: 48px;
+  align-items: center;
+  justify-content: space-between;
+  gap: 16px;
+  margin-top: 4px;
+  padding: 10px 14px;
+  color: #344054;
+  text-align: left;
+  background: #fff;
+  border: 1px solid #e2e5ec;
+  border-radius: 8px;
+  cursor: pointer;
+  transition: border-color 0.18s ease, background-color 0.18s ease;
+
+  &:hover {
+    background: #f8f7ff;
+    border-color: #b7b2f7;
+  }
+
+  > span {
+    display: grid;
+    gap: 3px;
+  }
+
+  strong {
+    font-size: 13px;
+    font-weight: 600;
+  }
+
+  small {
+    color: #8a93a3;
+    font-size: 12px;
+    line-height: 1.5;
+  }
+
+  i {
+    flex: 0 0 auto;
+    color: #7b8495;
+    font-size: 12px;
+  }
+}
+
+.advanced-settings-panel {
+  display: grid;
+  gap: 16px;
+  margin-top: -8px;
+  padding: 16px;
+  background: #fff;
+  border: 1px solid #e2e5ec;
+  border-radius: 8px;
+}
+
+.advanced-settings-grid {
+  display: grid;
+  grid-template-columns: repeat(2, minmax(0, 1fr));
+  gap: 12px;
+}
+
+.config-field {
+  display: grid;
+  align-content: start;
+  gap: 8px;
+  min-width: 0;
+  padding: 14px;
+  color: #344054;
+  background: #fff;
+  border: 1px solid #e2e5ec;
+  border-radius: 8px;
+
+  > small {
+    color: #8a93a3;
+    font-size: 12px;
+    line-height: 1.5;
+  }
+
+  :deep(.el-slider) {
+    margin: 0 8px;
+  }
+}
+
+.config-field-label {
+  font-size: 13px;
+  font-weight: 600;
+}
+
+.unit-input {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  color: #7b8495;
+  font-size: 12px;
+
+  :deep(.el-input-number) {
+    width: 100%;
+  }
+}
+
+.json-mode-row {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 20px;
+  padding: 14px;
+  background: #fff;
+  border: 1px solid #e2e5ec;
+  border-radius: 8px;
+}
+
+.quality-switch-row { display: flex; align-items: center; justify-content: space-between; gap: 20px; }
+.quality-rule-panel { margin-top: 12px; padding-top: 12px; border-top: 1px solid #e7eaf0; }
+.quality-rule-list { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); gap: 10px; }
+.quality-rule-list :deep(.el-checkbox) {
+  min-height: 62px; height: auto; margin-right: 0; padding: 10px 12px;
+  align-items: flex-start; border: 1px solid #e2e5ec; border-radius: 7px; background: #fff;
+}
+.quality-rule-list :deep(.el-checkbox__input) { margin-top: 3px; }
+.quality-rule-list :deep(.el-checkbox__label) { min-width: 0; padding-left: 9px; white-space: normal; }
+.rule-copy { display: grid; gap: 3px; }
+.rule-copy strong { color: #344054; font-size: 12px; font-weight: 600; }
+.rule-copy small { color: #8a93a3; font-size: 11px; line-height: 1.5; }
+
+.minimum-length-field {
+  margin-top: 12px; display: grid; grid-template-columns: 110px 220px minmax(0, 1fr);
+  align-items: center; gap: 12px; color: #475467; font-size: 12px;
+}
+.minimum-length-input { display: flex; align-items: center; gap: 8px; }
+.minimum-length-input :deep(.el-input-number) { width: 180px; }
+.minimum-length-field > small { color: #8a93a3; }
+.quality-rule-warning { margin: 10px 0 0; color: #dc2626; font-size: 12px; }
+.generation-validation-message { margin: 10px 0 0; color: #dc2626; font-size: 12px; }
+
+@media (max-width: 900px) {
+  .advanced-settings-grid { grid-template-columns: 1fr; }
+  .quality-rule-list { grid-template-columns: 1fr; }
+  .minimum-length-field { grid-template-columns: 1fr; }
+}
+</style>

frontend/src/views/data-process/create/GenerationStep.vue

@@ -97,7 +97,7 @@ const emit = defineEmits<{
   align-items: center;
   justify-content: space-between;
   padding: 15px 17px;
-  background: #fbfcfe;
+  background: #fff;
   border-bottom: 1px solid #e8ebf0;
 
   strong {

frontend/src/views/data-process/create/PreviewCompareStep.vue

@@ -350,7 +350,7 @@ function lineRange(item: PreviewItem) {
   min-height: 52px;
   padding: 0 15px;
   color: #313949;
-  background: #fbfcfe;
+  background: #fff;
   border-bottom: 1px solid #e8ebf0;
   font-size: 13px;
 

frontend/src/views/data-process/create/ResultEditorStep.vue

@@ -143,7 +143,7 @@ function selectRelative(offset: number) {
   min-height: 52px;
   padding: 0 15px;
   color: #344054;
-  background: #fbfcfe;
+  background: #fff;
   border-bottom: 1px solid #e8ebf0;
   font-size: 13px;
 

frontend/src/views/data-process/create/SourceUploadStep.vue

@@ -397,7 +397,7 @@ function formatSize(size: number) {
   width: 100%;
   min-height: 154px;
   padding: 32px 20px;
-  background: #fbfcfe;
+  background: #fff;
   border-color: #dfe3ea;
   transition: border-color 0.18s ease, background-color 0.18s ease;
 
@@ -435,7 +435,7 @@ function formatSize(size: number) {
   padding: 10px 14px;
   color: #5f6878;
   font-size: 12px;
-  background: #fbfcfe;
+  background: #fff;
   border-bottom: 1px solid #edf0f5;
 }
 

frontend/src/views/data-process/create/TaskSetupStep.vue

@@ -4,12 +4,15 @@ import type { FormInstance, FormRules } from 'element-plus'
 import type {
   ChunkMethod,
   DatasetSplitOptions,
+  GenerationControlOptions,
   PreprocessOption,
   ProcessType,
   StructuredProcessOptions,
   UnstructuredPreprocessOption,
   UnstructuredProcessOptions,
 } from './types'
+import GenerationOptionsPanel from './GenerationOptionsPanel.vue'
+import type { ModelItem } from '@/types'
 
 const props = defineProps<{
   name: string
@@ -17,6 +20,7 @@ const props = defineProps<{
   processType: ProcessType
   structuredOptions: StructuredProcessOptions
   unstructuredOptions: UnstructuredProcessOptions
+  generationModels: ModelItem[]
 }>()
 
 const emit = defineEmits<{
@@ -63,6 +67,10 @@ function updateStructuredField<K extends keyof StructuredProcessOptions>(
   emit('update:structuredOptions', { ...props.structuredOptions, [field]: value })
 }
 
+function updateStructuredGenerationOptions(value: GenerationControlOptions) {
+  emit('update:structuredOptions', { ...props.structuredOptions, ...value })
+}
+
 function updatePreprocessOptions(value: Array<string | number | boolean>) {
   const allowedValues = new Set(PREPROCESS_OPTIONS.map((option) => option.value))
   const preprocessOptions = value.filter(
@@ -89,6 +97,10 @@ function updateUnstructuredField<K extends keyof UnstructuredProcessOptions>(
   emit('update:unstructuredOptions', { ...props.unstructuredOptions, [field]: value })
 }
 
+function updateUnstructuredGenerationOptions(value: GenerationControlOptions) {
+  emit('update:unstructuredOptions', { ...props.unstructuredOptions, ...value })
+}
+
 function updateSmartPreprocess(value: string | number | boolean) {
   const enabled = Boolean(value)
   const remainingOptions = props.unstructuredOptions.preprocessOptions.filter(
@@ -143,6 +155,7 @@ function updateUnstructuredDatasetSplit(field: keyof DatasetSplitOptions, value:
 
 const formRef = ref<FormInstance>()
 const advancedChunkSettingsOpen = ref(props.unstructuredOptions.chunkMethod === 'custom')
+const validationAttempted = ref(false)
 const formModel = computed(() => ({
   name: props.name,
   processType: props.processType,
@@ -172,6 +185,25 @@ const preserveSpecialContentEnabled = computed(() => (
   && props.unstructuredOptions.preserveLists
 ))
 
+const activeGenerationOptions = computed<GenerationControlOptions | null>(() => {
+  if (props.processType === 'structured') return props.structuredOptions
+  if (props.processType === 'unstructured') return props.unstructuredOptions
+  return null
+})
+
+const generationValidationMessage = computed(() => {
+  const options = activeGenerationOptions.value
+  if (!options) return ''
+  if (options.generationModelId === '') return '请选择数据生成模型'
+  if (options.qualityFilterEnabled && !options.filterLowQuality && !options.filterShortContent) {
+    return '开启质量筛选后，请至少选择一项筛选规则'
+  }
+  if (options.qualityFilterEnabled && options.filterShortContent && options.minOutputLength < 1) {
+    return '最少字数必须大于 0'
+  }
+  return ''
+})
+
 const chunkValidationMessage = computed(() => {
   if (props.unstructuredOptions.chunkOverlap >= props.unstructuredOptions.chunkSize) {
     return '重叠长度必须小于切片长度'
@@ -208,6 +240,7 @@ const rules: FormRules = {
 }
 
 async function validate() {
+  validationAttempted.value = true
   if (!formRef.value) return false
   try {
     await formRef.value.validate()
@@ -219,6 +252,7 @@ async function validate() {
         return false
       }
     }
+    if (generationValidationMessage.value) return false
     return true
   } catch {
     return false
@@ -344,6 +378,13 @@ defineExpose({ validate })
             </div>
           </div>
           <div class="generation-option-list">
+            <GenerationOptionsPanel
+              :options="structuredOptions"
+              :models="generationModels"
+              section="quality"
+              :validation-message="validationAttempted ? generationValidationMessage : ''"
+              @update:options="updateStructuredGenerationOptions"
+            />
             <div class="generation-option-row">
               <div class="generation-option-copy">
                 <strong>语义丰富表达</strong>
@@ -435,6 +476,24 @@ defineExpose({ validate })
             </div>
           </div>
         </div>
+
+        <div class="form-section model-options-section">
+          <div class="section-title-row">
+            <div>
+              <h3>大模型</h3>
+              <p>选择数据生成模型，并设置模型输出内容的要求</p>
+            </div>
+          </div>
+          <div class="generation-option-list">
+            <GenerationOptionsPanel
+              :options="structuredOptions"
+              :models="generationModels"
+              section="model"
+              :validation-message="validationAttempted ? generationValidationMessage : ''"
+              @update:options="updateStructuredGenerationOptions"
+            />
+          </div>
+        </div>
       </template>
 
       <template v-if="processType === 'unstructured'">
@@ -625,6 +684,13 @@ defineExpose({ validate })
             </div>
           </div>
           <div class="generation-option-list">
+            <GenerationOptionsPanel
+              :options="unstructuredOptions"
+              :models="generationModels"
+              section="quality"
+              :validation-message="validationAttempted ? generationValidationMessage : ''"
+              @update:options="updateUnstructuredGenerationOptions"
+            />
             <div class="generation-option-row">
               <div class="generation-option-copy">
                 <strong>语义丰富表达</strong>
@@ -723,6 +789,24 @@ defineExpose({ validate })
             </div>
           </div>
         </div>
+
+        <div class="form-section model-options-section">
+          <div class="section-title-row">
+            <div>
+              <h3>大模型</h3>
+              <p>选择数据生成模型，并设置模型输出内容的要求</p>
+            </div>
+          </div>
+          <div class="generation-option-list">
+            <GenerationOptionsPanel
+              :options="unstructuredOptions"
+              :models="generationModels"
+              section="model"
+              :validation-message="validationAttempted ? generationValidationMessage : ''"
+              @update:options="updateUnstructuredGenerationOptions"
+            />
+          </div>
+        </div>
       </template>
 
     </el-form>
@@ -843,7 +927,7 @@ defineExpose({ validate })
   min-width: 0;
   padding: 14px;
   color: #344054;
-  background: #fbfcfe;
+  background: #fff;
   border: 1px solid #e2e5ec;
   border-radius: 8px;
 
@@ -902,7 +986,7 @@ defineExpose({ validate })
   padding: 10px 14px;
   color: #344054;
   text-align: left;
-  background: #fbfcfe;
+  background: #fff;
   border: 1px solid #e2e5ec;
   border-radius: 8px;
   cursor: pointer;
@@ -946,7 +1030,7 @@ defineExpose({ validate })
   gap: 12px;
   margin-top: 8px;
   padding: 12px;
-  background: #fbfcfe;
+  background: #fff;
   border: 1px solid #e2e5ec;
   border-radius: 8px;
 
@@ -1154,3 +1238,4 @@ defineExpose({ validate })
 
 }
 </style>
+  GenerationControlOptions,

frontend/src/views/data-process/create/previewModel.ts

@@ -501,7 +501,11 @@ export function createResults(
     ? Math.min(3, finiteInteger(options.qaPairsPerChunk, 1, 1))
     : Math.min(5, finiteInteger(options?.qaPairsPerRow, 1, 1))
 
-  return items.flatMap((item, index) => {
+  const generatedResults = items.flatMap((item, index) => {
+    if (options?.qualityFilterEnabled && options.filterLowQuality) {
+      if (item.status === 'invalid' || !item.editedContent.trim()) return []
+    }
+
     const [firstLine = '', ...rest] = item.editedContent.split('\n')
     const output = rest.join('\n').trim() || item.editedContent.trim()
     const baseInstruction = firstLine.replace(/^问[:：]\s*/, '').trim() || `数据条目 ${index + 1}`
@@ -525,4 +529,12 @@ export function createResults(
       }
     })
   })
+
+  if (!options?.qualityFilterEnabled) return generatedResults
+
+  return generatedResults.filter((result) => {
+    if (options.filterLowQuality && (!result.instruction.trim() || !result.output.trim())) return false
+    if (options.filterShortContent && result.output.trim().length < options.minOutputLength) return false
+    return true
+  })
 }

frontend/src/views/data-process/create/types.ts

@@ -16,7 +16,19 @@ export interface DatasetSplitOptions {
   test: number
 }
 
-export interface StructuredProcessOptions {
+export interface GenerationControlOptions {
+  generationModelId: string | number | ''
+  generationPrompt: string
+  temperature: number
+  maxTokens: number
+  jsonMode: boolean
+  qualityFilterEnabled: boolean
+  filterLowQuality: boolean
+  filterShortContent: boolean
+  minOutputLength: number
+}
+
+export interface StructuredProcessOptions extends GenerationControlOptions {
   preprocessOptions: PreprocessOption[]
   semanticEnrichment: boolean
   qaPairsPerRow: number
@@ -34,7 +46,7 @@ export type UnstructuredPreprocessOption =
 
 export type ChunkMethod = 'semantic' | 'heading' | 'fixed' | 'custom'
 
-export interface UnstructuredProcessOptions {
+export interface UnstructuredProcessOptions extends GenerationControlOptions {
   preprocessOptions: UnstructuredPreprocessOption[]
   chunkMethod: ChunkMethod
   chunkSize: number