feat: 数据处理向导新增模型配置步骤

StepId 新增 model 步骤，DataProcessCreateView 增加模型选择与生成参数编排，各选项面板与样式适配新步骤流程，回归脚本补充断言。
2026-07-13 17:12:49 +08:00
parent 42f2efb449
commit 762d48d090
9 changed files with 257 additions and 121 deletions
--- a/frontend/scripts/regression-data-process-wizard.mjs
+++ b/frontend/scripts/regression-data-process-wizard.mjs
@@ -41,15 +41,20 @@ assert.match(viewSource, /onBeforeRouteLeave\(async \(\) =>/, '路由离开确
 assert.doesNotMatch(viewSource, /window\.confirm|ElMessageBox/, '创建页仍在使用系统或 Element Plus 确认框')

 assert.match(viewSource, /const WIZARD_STEPS = \[/, '向导步骤尚未改为固定常量')
-for (const title of ['创建任务', '上传文件', '数据预览', '开始生成', '结果编辑与保存']) {
+for (const title of ['创建任务', '大模型选择', '上传文件', '数据预览', '开始生成', '结果编辑与保存']) {
  assert.ok(viewSource.includes(`title: '${title}'`), `缺少固定步骤：${title}`)
 }
 assert.match(
  viewSource,
-  /\{ id: 'create',[\s\S]*?\{ id: 'upload',[\s\S]*?\{ id: 'preview',[\s\S]*?\{ id: 'generate',[\s\S]*?\{ id: 'results'/,
-  '五步向导顺序必须为创建任务、上传文件、数据预览、开始生成、结果编辑与保存',
+  /\{ id: 'create',[\s\S]*?\{ id: 'model',[\s\S]*?\{ id: 'upload',[\s\S]*?\{ id: 'preview',[\s\S]*?\{ id: 'generate',[\s\S]*?\{ id: 'results'/,
+  '六步向导顺序必须为创建任务、大模型选择、上传文件、数据预览、开始生成、结果编辑与保存',
 )
 assert.doesNotMatch(viewSource, /steps\s*=\s*computed|all\.filter/, '步骤仍根据处理类型动态增减')
+assert.match(
+  viewStyleSource,
+  /@media \(max-width: 1100px\)[\s\S]*?\.step-title\s*\{[\s\S]*?display:\s*none[\s\S]*?\.step-item\.is-active \.step-title\s*\{[\s\S]*?display:\s*block/,
+  '六步向导在中等宽度下没有收起非当前步骤标题',
+)
 assert.match(draftSource, /localStorage\.setItem\(DATA_PROCESS_DRAFT_STORAGE_KEY/, '草稿没有持久化')
 assert.match(draftSource, /localStorage\.getItem\(DATA_PROCESS_DRAFT_STORAGE_KEY\)/, '草稿没有恢复读取')
 assert.match(viewSource, /restoreDraft\(\)/, '页面没有恢复草稿')
@@ -57,6 +62,7 @@ assert.ok(viewSource.split('\n').length < 800, 'DataProcessCreateView 拆分后

 const expectedComponents = [
  'TaskSetupStep.vue',
+  'ModelSelectionStep.vue',
  'SourceUploadStep.vue',
  'PreviewCompareStep.vue',
  'GenerationStep.vue',
@@ -82,7 +88,7 @@ for (const field of ['sourceStart', 'sourceEnd', 'originalContent', 'editedConte
  assert.ok(typesSource.includes(field), `PreviewItem 缺少字段：${field}`)
 }
 assert.match(typesSource, /sourceFileId/, 'PreviewItem 缺少来源文件标识')
-assert.match(typesSource, /export type StepId = 'create' \| 'upload' \| 'preview' \| 'generate' \| 'results'/, '步骤类型缺少独立上传步骤')
+assert.match(typesSource, /export type StepId = 'create' \| 'model' \| 'upload' \| 'preview' \| 'generate' \| 'results'/, '步骤类型缺少独立大模型选择步骤')
 assert.match(modelSource, /export function buildPreviewItems/, '缺少切片来源映射生成函数')
 assert.match(modelSource, /export function sourceLines/, '缺少源文件行偏移生成函数')
 assert.match(modelSource, /sourceFileId/, '切片生成没有写入来源文件标识')
@@ -102,10 +108,10 @@ for (const marker of [
  'preview-editor',
  'scrollIntoView',
 ]) {
-  assert.ok(previewSource.includes(marker), `第三步缺少结构或行为：${marker}`)
+  assert.ok(previewSource.includes(marker), `第四步缺少结构或行为：${marker}`)
 }
-assert.match(previewSource, /sourceStart/, '第三步未使用来源起始偏移')
-assert.match(previewSource, /sourceEnd/, '第三步未使用来源结束偏移')
+assert.match(previewSource, /sourceStart/, '第四步未使用来源起始偏移')
+assert.match(previewSource, /sourceEnd/, '第四步未使用来源结束偏移')
 assert.match(previewSource, /filterable/, '文件选择器必须可搜索')
 assert.match(previewSource, /当前文件/, '预览缺少当前文件切换器')
 assert.doesNotMatch(previewSource, /located-badge|sync-label|已定位到/, '源文件栏不应显示冗余定位提示')
@@ -130,13 +136,14 @@ assert.match(previewSource, />保存修改<\/el-button>/, '编辑器缺少保存
 assert.match(previewSource, /\.editor-actions\s*\{[\s\S]*?justify-content:\s*flex-end/, '取消和保存按钮必须在编辑器右侧对齐')
 assert.doesNotMatch(previewSource, /item-token|item-status|modifiedOnly|仅看已修改/, '切片列表不应再显示 Token 或修改状态')
 assert.match(previewSource, /\.preview-editor\s*\{[\s\S]*?flex:\s*1 1 auto[\s\S]*?overflow-y:\s*auto/, '编辑模式必须占据右侧剩余区域并可滚动')
-assert.match(previewSource, /@media \(max-width: 900px\)/, '第三步缺少窄屏上下布局')
+assert.match(previewSource, /@media \(max-width: 900px\)/, '第四步缺少窄屏上下布局')

 const taskSetupPath = path.join(createDir, 'TaskSetupStep.vue')
 const structuredOptionsPath = path.join(createDir, 'StructuredOptionsPanel.vue')
 const unstructuredOptionsPath = path.join(createDir, 'UnstructuredOptionsPanel.vue')
 const datasetSplitEditorPath = path.join(createDir, 'DatasetSplitEditor.vue')
 const generationOptionsPath = path.join(createDir, 'GenerationOptionsPanel.vue')
+const modelSelectionPath = path.join(createDir, 'ModelSelectionStep.vue')
 const sourceUploadPath = path.join(createDir, 'SourceUploadStep.vue')
 const [
  taskSetupSource,
@@ -144,6 +151,7 @@ const [
  unstructuredOptionsSource,
  datasetSplitEditorSource,
  generationControlSource,
+  modelSelectionSource,
  sourceUploadSource,
 ] = await Promise.all([
  readFile(taskSetupPath, 'utf8'),
@@ -151,6 +159,7 @@ const [
  readFile(unstructuredOptionsPath, 'utf8'),
  readFile(datasetSplitEditorPath, 'utf8'),
  readFile(generationOptionsPath, 'utf8'),
+  readFile(modelSelectionPath, 'utf8'),
  readFile(sourceUploadPath, 'utf8'),
 ])
 const taskSetupFeatureSource = [
@@ -172,20 +181,29 @@ assert.match(unstructuredOptionsSource, /<DatasetSplitEditor/, '非结构化选
 for (const marker of ['<el-upload', '源数据上传', '数据源配置', 'uploadedFiles']) {
  assert.ok(!taskSetupFeatureSource.includes(marker), `第一步仍包含上传职责：${marker}`)
 }
-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, /<ModelSelectionStep\s+[\s\S]*?v-else-if="currentStepId === 'model'"/, '第二步没有挂载独立大模型选择组件')
+assert.match(viewSource, /<SourceUploadStep\s+[\s\S]*?v-else-if="currentStepId === 'upload'"/, '第三步没有挂载独立上传组件')
+assert.match(viewSource, /if \(currentStepId\.value === 'create'\) return '继续：选择大模型'/, '第一步主按钮没有指向大模型选择')
+assert.match(viewSource, /if \(currentStepId\.value === 'model'\) return '继续：上传文件'/, '第二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'/, '第三步主按钮没有指向数据预览')
 assert.match(draftSource, /DATA_PROCESS_DRAFT_SCHEMA_VERSION = 6/, '安全草稿格式必须升级到 v6')

 const nextFromCreateStart = viewSource.indexOf('async function nextFromCreate()')
-const nextFromUploadStart = viewSource.indexOf('function nextFromUpload()', nextFromCreateStart)
+const nextFromModelStart = viewSource.indexOf('async function nextFromModel()', nextFromCreateStart)
+const nextFromUploadStart = viewSource.indexOf('function nextFromUpload()', nextFromModelStart)
 const selectPreviewFileStart = viewSource.indexOf('function selectPreviewFile(', nextFromUploadStart)
-assert.ok(nextFromCreateStart >= 0 && nextFromUploadStart > nextFromCreateStart, '缺少创建步骤与上传步骤的独立跳转函数')
-const nextFromCreateSource = viewSource.slice(nextFromCreateStart, nextFromUploadStart)
+assert.ok(
+  nextFromCreateStart >= 0 && nextFromModelStart > nextFromCreateStart && nextFromUploadStart > nextFromModelStart,
+  '缺少创建、大模型选择与上传步骤的独立跳转函数',
+)
+const nextFromCreateSource = viewSource.slice(nextFromCreateStart, nextFromModelStart)
+const nextFromModelSource = viewSource.slice(nextFromModelStart, nextFromUploadStart)
 const nextFromUploadSource = viewSource.slice(nextFromUploadStart, selectPreviewFileStart)
 assert.match(nextFromCreateSource, /taskSetupRef\.value\?\.validate\(\)/, '创建步骤继续前没有校验任务配置')
-assert.match(nextFromCreateSource, /goToStep\('upload'\)/, '创建步骤校验通过后没有进入上传文件')
+assert.match(nextFromCreateSource, /goToStep\('model'\)/, '创建步骤校验通过后没有进入大模型选择')
 assert.doesNotMatch(nextFromCreateSource, /uploadedFiles|buildPreviewItems/, '创建步骤仍在校验文件或提前生成预览')
+assert.match(nextFromModelSource, /modelSelectionRef\.value\?\.validate\(\)/, '大模型选择步骤继续前没有校验模型配置')
+assert.match(nextFromModelSource, /goToStep\('upload'\)/, '大模型选择完成后没有进入上传文件')
 assert.match(nextFromUploadSource, /uploadedFiles\.value\.length === 0/, '上传步骤继续前没有校验源数据')
 assert.match(nextFromUploadSource, /buildPreviewItems\(/, '上传步骤没有在进入预览前生成预览数据')
 assert.match(nextFromUploadSource, /goToStep\('preview'\)/, '上传步骤完成后没有进入数据预览')
@@ -217,8 +235,6 @@ for (const option of [
  assert.ok(structuredOptionsSource.includes(option), `结构化预处理缺少选项：${option}`)
 }
 assert.ok(structuredOptionsSource.includes('生成选项'), '结构化配置缺少生成选项分类')
-assert.ok(structuredOptionsSource.includes('语义丰富表达'), '生成选项缺少语义丰富表达开关')
-assert.ok(structuredOptionsSource.includes('使用大模型将问答表述得更自然、柔和'), '语义丰富表达缺少辅助说明')
 for (const splitName of ['训练集', '验证集', '测试集']) {
  assert.ok(datasetSplitEditorSource.includes(splitName), `生成选项缺少数据集划分：${splitName}`)
 }
@@ -232,7 +248,6 @@ for (const splitField of ['train', 'validation', 'test']) {
    `数据集划分字段 ${splitField} 缺少 0～100 的整数限制`,
  )
 }
-assert.match(structuredOptionsSource, /<el-switch[\s\S]*options\.semanticEnrichment/, '语义丰富表达必须使用开关控件')
 assert.match(structuredOptionsSource, /<el-input-number[\s\S]*options\.qaPairsPerRow[\s\S]*:min="1"[\s\S]*:max="5"/, '每行生成数量必须限制在 1 到 5')
 assert.match(viewSource, /const structuredOptions = ref<StructuredProcessOptions>/, '父页面缺少结构化配置状态')
 assert.match(stateSource, /datasetSplit:\s*\{ train: 80, validation: 10, test: 10 \}/, '数据集划分默认值必须为 80/10/10')
@@ -252,12 +267,16 @@ for (const field of [
  assert.ok(typesSource.includes(field), `生成控制配置缺少字段：${field}`)
  assert.ok(implementationSource.includes(field), `父页面默认值或草稿状态缺少字段：${field}`)
 }
-assert.match(structuredOptionsSource, /GenerationOptionsPanel/, '结构化生成选项没有复用统一的大模型与质量筛选组件')
-assert.match(unstructuredOptionsSource, /GenerationOptionsPanel/, '非结构化生成选项没有复用统一的大模型与质量筛选组件')
+assert.match(structuredOptionsSource, /GenerationOptionsPanel/, '结构化生成选项没有复用统一的质量筛选组件')
+assert.match(unstructuredOptionsSource, /GenerationOptionsPanel/, '非结构化生成选项没有复用统一的质量筛选组件')
 assert.match(structuredOptionsSource, /:options="options"/, '结构化生成选项未接入统一配置组件')
 assert.match(unstructuredOptionsSource, /:options="options"/, '非结构化生成选项未接入统一配置组件')
-assert.match(taskSetupFeatureSource, /<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(taskSetupFeatureSource, /section="model"/, '独立大模型分类没有挂载模型配置')
+assert.doesNotMatch(taskSetupFeatureSource, /<h3>大模型<\/h3>|section="model"/, '第一步不应继续承载大模型配置')
+assert.match(modelSelectionSource, /<h3[^>]*>大模型选择<\/h3>/, '独立步骤缺少大模型选择标题')
+assert.match(modelSelectionSource, /section="model"/, '独立步骤没有挂载模型配置')
+assert.match(modelSelectionSource, /defineExpose\(\{ validate \}\)/, '独立大模型选择步骤没有暴露继续前校验')
+assert.match(modelSelectionSource, /class="form-section"/, '大模型选择步骤没有沿用第一步的通栏表单分区')
+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']) {
@@ -267,15 +286,21 @@ assert.match(generationControlSource, /filterable/, '数据生成模型下拉必
 assert.match(generationControlSource, /maxlength="500"/, '默认提示语缺少合理的长度限制')
 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')
+assert.match(
+  generationControlSource,
+  /\.model-config-group \.advanced-settings-grid\s*\{[\s\S]*?grid-template-columns:\s*1fr/,
+  '大模型高级参数没有改为与第一步一致的纵向布局',
+)
 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.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(taskSetupSource, /qualityValidationMessage/, '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(viewSource, /<ModelSelectionStep[\s\S]*?:models="generationModels"/, '创建页没有向独立大模型选择步骤传递模型列表')

 assert.match(typesSource, /export interface UnstructuredProcessOptions/, '缺少非结构化处理选项类型')
 for (const field of [
@@ -312,11 +337,7 @@ for (const method of ['自动语义切分', '按标题和段落', '按固定长
 for (const label of ['切片长度', '重叠长度', '最小切片长度', '保护表格、代码和列表']) {
  assert.ok(unstructuredOptionsSource.includes(label), `切分选项缺少配置：${label}`)
 }
-assert.ok(unstructuredOptionsSource.includes('高级设置'), '切分选项缺少渐进式高级设置入口')
-assert.match(unstructuredOptionsSource, /:aria-expanded="advancedChunkSettingsOpen"/, '高级设置入口缺少展开状态的可访问性标记')
-assert.match(unstructuredOptionsSource, /v-if="advancedChunkSettingsOpen"/, '高级设置内容没有默认收起')
-assert.match(unstructuredOptionsSource, /watch\(\(\) => props\.options\.chunkMethod,[\s\S]*?method === 'custom'[\s\S]*?advancedChunkSettingsOpen\.value = true/, '选择自定义分隔符时没有自动展开高级设置')
-assert.match(unstructuredOptionsSource, /function revealValidation\(\)[\s\S]*?advancedChunkSettingsOpen\.value = true/, '非结构化面板没有暴露高级校验定位能力')
+assert.doesNotMatch(unstructuredOptionsSource, /advancedChunkSettingsOpen|>高级设置</, '切分核心参数不应再隐藏在高级设置中')
 assert.match(taskSetupSource, /if \(chunkValidationMessage\.value\) \{[\s\S]*?revealValidation\(\)[\s\S]*?return false/, '高级切分配置校验失败时没有重新展开定位')
 assert.match(unstructuredOptionsSource, /const preserveSpecialContentEnabled = computed/, '特殊内容保护没有合并为单一开关')
 assert.match(unstructuredOptionsSource, /function updateSpecialContentProtection/, '特殊内容保护开关缺少更新逻辑')
@@ -324,7 +345,7 @@ assert.match(unstructuredOptionsSource, /options\.chunkSize[\s\S]*?:min="200"[\s
 assert.match(unstructuredOptionsSource, /options\.chunkOverlap[\s\S]*?:min="0"[\s\S]*?:max="500"/, '重叠长度必须限制在 0 到 500 Token')
 assert.match(unstructuredOptionsSource, /options\.minChunkSize[\s\S]*?:min="20"[\s\S]*?:max="500"/, '最小切片长度必须限制在 20 到 500 Token')

-for (const label of ['语义丰富表达', '每个切片生成数量', '数据集划分']) {
+for (const label of ['每个切片生成数量', '数据集划分']) {
  assert.ok(taskSetupFeatureSource.includes(label), `非结构化生成选项缺少：${label}`)
 }
 for (const removedLabel of ['上下文范围', '问题类型', '跳过无法回答的内容']) {
@@ -862,4 +883,4 @@ assert.match(
 )
 assert.match(previewSource, /height:\s*clamp\(560px,\s*calc\(100vh - 370px\),\s*720px\)/, '对照预览高度不足以展示切片正文')

-console.log('数据处理五步向导回归检查通过')
+console.log('数据处理六步向导回归检查通过')