feat(data-process): 支持思维链输出类型

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

@@ -86,6 +86,8 @@ assert.doesNotMatch(detailSource, /^\s*max-width:\s*\d+px/m, '详情页不应使
 assert.match(detailSource, /!\/\(\?:password\|secret\|token\|api_key\)\/i\.test\(key\)/, '处理配置没有过滤敏感凭据字段')
 assert.match(detailSource, /key !== 'generation_model_snapshot'/, '处理配置仍直接展示内部模型快照')
 assert.match(detailSource, /preprocessOptionLabelMap/, '处理配置没有把预处理内部枚举转换为中文')
+assert.match(detailSource, /output_type:\s*'输出类型'/, '处理配置没有显示输出类型名称')
+assert.match(detailSource, /value === 'reasoning' \? '思维链回答' : '标准回答'/, '处理配置没有转换输出类型枚举')
 assert.match(detailSource, /const parsed = typeof value === 'number' \? value : Number\(value\)/, '详情页不兼容 PostgreSQL 数字字符串')
 assert.match(detailSource, /new Date\(startTime\.value\)\.getTime\(\)/, '详情页没有在后端耗时缺失时按开始、完成时间回算')
 assert.match(detailSource, /outputCount \+ numeric\(detail\.value\?\.filtered_count\)/, '结果保留率没有使用输出和过滤结果的同口径分母')

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

@@ -452,6 +452,7 @@ assert.match(apiSource, /regenerateDataProcessTask[\s\S]*?\/regenerate`/, '重
 for (const field of [
   'generationModelId',
   'generationPrompt',
+  'outputType',
   'qualityFilterEnabled',
   'filterLowQuality',
   'filterShortContent',
@@ -472,11 +473,14 @@ assert.match(modelSelectionSource, /class="form-section"/, '大模型选择步
 assert.doesNotMatch(modelSelectionSource, /max-width:\s*980px/, '大模型选择步骤不应使用比第一步更窄的固定内容宽度')
 assert.match(taskSetupFeatureSource, /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']) {
+for (const label of ['大模型', '数据生成模型', '默认提示语', '输出类型', '标准回答', '思维链回答', '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, /aria-label="输出类型"/, '输出类型选项缺少可访问名称')
+assert.match(generationControlSource, /&lt;think&gt;推理过程&lt;\/think&gt;/, '思维链选项没有说明最终保存格式')
+assert.match(generationControlSource, /\.output-type-options[\s\S]*?min-height:\s*44px/, '输出类型选项的点击区域不足 44px')
 assert.match(generationControlSource, /\.model-field\s*\{[\s\S]*?display:\s*flex[\s\S]*?flex-direction:\s*column/, '大模型字段没有使用稳定的纵向表单布局')
 assert.match(generationControlSource, /\.generation-config-group\s*\{[\s\S]*?border:\s*1px solid #e2e5ec/, '大模型配置没有保留统一配置面板边框')
 assert.match(generationControlSource, /\.model-config-group\s*\{[\s\S]*?padding:\s*0[\s\S]*?border:\s*0/, '独立大模型步骤仍存在嵌套卡片挤压')
@@ -487,6 +491,7 @@ assert.match(
 )
 assert.match(stateSource, /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((stateSource.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.equal((stateSource.match(/outputType:\s*'standard'/g) || []).length, 2, '结构化与非结构化任务应默认生成标准回答')
 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')
@@ -659,6 +664,7 @@ for (const [backendField, frontendField] of [
   ['semantic_enrichment', 'semanticEnrichment'],
   ['generation_model_id', 'generationModelId'],
   ['generation_prompt', 'generationPrompt'],
+  ['output_type', 'outputType'],
   ['temperature', 'temperature'],
   ['max_tokens', 'maxTokens'],
   ['json_mode', 'jsonMode'],
@@ -702,6 +708,7 @@ assert.match(stateSource, /Object\.prototype\.hasOwnProperty\.call\(config, key\
 assert.match(stateSource, /Number\.isFinite\(value\) \? value : fallback/, '配置反向映射没有保留合法数字 0')
 assert.match(stateSource, /qaPairsPerRow:\s*normalizeQaPairsGenerationCount\([\s\S]*?qa_pairs_per_row[\s\S]*?defaults\.qaPairsPerRow/, '结构化生成数量回填没有按 1 到 50 归一化')
 assert.match(stateSource, /qaPairsPerChunk:\s*normalizeQaPairsGenerationCount\([\s\S]*?qa_pairs_per_chunk[\s\S]*?defaults\.qaPairsPerChunk/, '非结构化生成数量回填没有按 1 到 50 归一化')
+assert.match(stateSource, /outputType:\s*configValue\(config, 'output_type', defaults\.outputType\) === 'reasoning'[\s\S]*?\? 'reasoning'[\s\S]*?: 'standard'/, '输出类型没有从任务配置安全回填')
 assert.match(stateSource, /createStructuredOptionsFromConfig/, '结构化配置缺少后端到表单的反向映射')
 assert.match(stateSource, /createUnstructuredOptionsFromConfig/, '非结构化配置缺少后端到表单的反向映射')
 assert.match(stateSource, /configValue<unknown>\(config, 'preprocess_options', \[\]\)/, '历史任务缺少预处理配置时必须按后端空列表语义回填')
@@ -778,6 +785,7 @@ for (const field of [
   'datasetSplit',
   'generationModelId',
   'generationPrompt',
+  'outputType',
   'qualityFilterEnabled',
   'filterLowQuality',
   'filterShortContent',

frontend/src/types/dataProcess.ts

@@ -4,6 +4,7 @@ export type DataProcessStatus = 'pending' | 'running' | 'completed' | 'failed' |
 export type DataProcessType = 'structured' | 'unstructured' | 'external'
 export type DataProcessResultStatus = 'valid' | 'modified' | 'invalid'
 export type DataProcessSplit = 'train' | 'validation' | 'test'
+export type DataProcessOutputType = 'standard' | 'reasoning'
 
 export interface DataProcessPage<T> {
   items: T[]
@@ -20,6 +21,7 @@ export interface DataProcessDatasetSplit {
 
 export type DataProcessConfig = Record<string, unknown> & {
   dataset_split?: DataProcessDatasetSplit
+  output_type?: DataProcessOutputType
 }
 
 export interface DataProcessTask {

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

@@ -195,13 +195,13 @@ function toBackendConfig(): DataProcessConfig {
   const options = processType.value === 'unstructured'
     ? unstructuredOptions.value
     : structuredOptions.value
-
   const common = {
     preprocess_options: [...options.preprocessOptions],
     semantic_enrichment: options.semanticEnrichment,
     dataset_split: { ...options.datasetSplit },
     generation_model_id: options.generationModelId,
     generation_prompt: options.generationPrompt,
+    output_type: options.outputType,
     temperature: options.temperature,
     max_tokens: options.maxTokens,
     json_mode: options.jsonMode,

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

@@ -70,6 +70,7 @@ const configLabelMap: Record<string, string> = {
   dataset_split: '数据集划分',
   generation_model_id: '数据生成模型',
   generation_prompt: '生成提示语',
+  output_type: '输出类型',
   temperature: '生成温度',
   max_tokens: '最大输出长度',
   json_mode: 'JSON 输出',
@@ -245,6 +246,9 @@ function formatConfigValue(key: string, value: unknown) {
   if (key === 'chunk_method' && typeof value === 'string') {
     return chunkMethodLabelMap[value] || value
   }
+  if (key === 'output_type') {
+    return value === 'reasoning' ? '思维链回答' : '标准回答'
+  }
   if (key === 'dataset_split' && value && typeof value === 'object') {
     const split = value as Partial<DataProcessDatasetSplit>
     return `训练集 ${split.train ?? 0}% / 验证集 ${split.validation ?? 0}% / 测试集 ${split.test ?? 0}%`

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

@@ -30,6 +30,10 @@ function updateQualityRules(value: Array<string | number>) {
   })
 }
 
+function updateOutputType(value: string | number | boolean | undefined) {
+  updateField('outputType', value === 'reasoning' ? 'reasoning' : 'standard')
+}
+
 const selectedQualityRules = () => [
   props.options.filterLowQuality ? 'low_quality' : '',
   props.options.filterShortContent ? 'short_content' : '',
@@ -106,6 +110,28 @@ function modelMeta(model: ModelItem) {
       </div>
     </div>
 
+    <div class="model-field output-type-field">
+      <div class="field-copy">
+        <strong>输出类型</strong>
+        <small>控制答案是否包含可用于推理模型训练的思维链内容</small>
+      </div>
+      <el-radio-group
+        class="output-type-options"
+        :model-value="options.outputType"
+        aria-label="输出类型"
+        @update:model-value="updateOutputType"
+      >
+        <el-radio-button value="standard">标准回答</el-radio-button>
+        <el-radio-button value="reasoning">思维链回答</el-radio-button>
+      </el-radio-group>
+      <p class="output-type-hint">
+        <template v-if="options.outputType === 'reasoning'">
+          生成结果将按 <code>&lt;think&gt;推理过程&lt;/think&gt;</code> 加最终答案的格式保存。
+        </template>
+        <template v-else>仅保存最终答案，不包含推理过程。</template>
+      </p>
+    </div>
+
     <div class="advanced-settings-grid">
       <label class="config-field">
         <span class="config-field-label">生成温度 (Temperature)</span>
@@ -304,6 +330,31 @@ function modelMeta(model: ModelItem) {
   }
 }
 
+.output-type-options {
+  align-self: flex-start;
+
+  :deep(.el-radio-button__inner) {
+    min-height: 44px;
+    padding: 13px 22px;
+  }
+}
+
+.output-type-hint {
+  min-height: 20px;
+  margin: 0;
+  color: #667085;
+  font-size: 12px;
+  line-height: 1.6;
+
+  code {
+    padding: 2px 5px;
+    color: #5b50f2;
+    background: #f0f0ff;
+    border-radius: 4px;
+    font-family: inherit;
+  }
+}
+
 .advanced-settings-grid {
   display: grid;
   grid-template-columns: repeat(2, minmax(0, 1fr));

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

@@ -9,7 +9,7 @@ import type {
 } from './types'
 import { normalizeQaPairsGenerationCount } from './types'
 
-export 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。'
+export 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。仅输出符合所选输出类型和目标格式的内容，答案应事实清晰、语言自然，不要添加无关说明。'
 
 export function createDefaultStructuredOptions(): StructuredProcessOptions {
   return {
@@ -19,6 +19,7 @@ export function createDefaultStructuredOptions(): StructuredProcessOptions {
     datasetSplit: { train: 80, validation: 10, test: 10 },
     generationModelId: '',
     generationPrompt: DEFAULT_GENERATION_PROMPT,
+    outputType: 'standard',
     temperature: 0.7,
     maxTokens: 1024,
     jsonMode: false,
@@ -52,6 +53,7 @@ export function createDefaultUnstructuredOptions(): UnstructuredProcessOptions {
     datasetSplit: { train: 80, validation: 10, test: 10 },
     generationModelId: '',
     generationPrompt: DEFAULT_GENERATION_PROMPT,
+    outputType: 'standard',
     temperature: 0.7,
     maxTokens: 1024,
     jsonMode: false,
@@ -93,6 +95,9 @@ function generationOptionsFromConfig(
   return {
     generationModelId: configValue(config, 'generation_model_id', defaults.generationModelId),
     generationPrompt: String(configValue(config, 'generation_prompt', defaults.generationPrompt)),
+    outputType: configValue(config, 'output_type', defaults.outputType) === 'reasoning'
+      ? 'reasoning'
+      : 'standard',
     temperature: numberValue(config, 'temperature', defaults.temperature),
     maxTokens: numberValue(config, 'max_tokens', defaults.maxTokens),
     jsonMode: Boolean(configValue(config, 'json_mode', defaults.jsonMode)),
@@ -218,6 +223,7 @@ export function generationAffectingOptionsFor(
     datasetSplit: options.datasetSplit,
     generationModelId: options.generationModelId,
     generationPrompt: options.generationPrompt,
+    outputType: options.outputType,
     temperature: options.temperature,
     maxTokens: options.maxTokens,
     jsonMode: options.jsonMode,

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

@@ -1,4 +1,4 @@
-import type { DataProcessPreviewFileStatus } from '@/types/dataProcess'
+import type { DataProcessOutputType, DataProcessPreviewFileStatus } from '@/types/dataProcess'
 
 export type ProcessType = 'structured' | 'unstructured' | 'external'
 
@@ -32,6 +32,7 @@ export interface DatasetSplitOptions {
 export interface GenerationControlOptions {
   generationModelId: string | number | ''
   generationPrompt: string
+  outputType: DataProcessOutputType
   temperature: number
   maxTokens: number
   jsonMode: boolean