fix(pipeline): make data-analysis stage opt-in per agent (#1244)

The legacy in-pipeline DuckDB SQL data-analysis stage used to run on every
quick-answer RAG request whose retrieved chunks included a CSV/Excel file,
adding one extra LLM round-trip (~3s) to generate a SQL query that most
plain Q&A users never wanted. There was no way to disable it.

Introduce a per-agent DataAnalysisEnabled flag (default off), wire it
through PipelineRequest, and gate the DATA_ANALYSIS stage on it. Surface
the toggle in the agent editor for quick-answer agents with at least one
knowledge base attached.

Refs: https://github.com/Tencent/WeKnora/issues/1244
This commit is contained in:
wizardchen
2026-05-12 19:36:28 +08:00
committed by lyingbug
parent 86b05d923e
commit 3f9b09e306
9 changed files with 49 additions and 1 deletions
+4
View File
@@ -4007,6 +4007,10 @@ export default {
boostLabel: 'FAQ Score Boost',
boostDesc: 'Multiply FAQ relevance scores by this factor to rank them higher',
},
dataAnalysis: {
enableLabel: 'Enable Tabular Data Analysis',
enableDesc: 'When the retrieved chunks come from a CSV/Excel file, ask the LLM to generate a DuckDB SQL query before answering. This adds one extra LLM call and several seconds of latency, so only enable it when you actually need SQL-style analysis.',
},
fallback: {
fixed: 'Fixed Response',
model: 'Model Generated',
+4
View File
@@ -4070,6 +4070,10 @@ export default {
boostLabel: 'FAQ 점수 가중치',
boostDesc: 'FAQ 관련성 점수에 이 계수를 곱하여 순위를 높입니다',
},
dataAnalysis: {
enableLabel: '표 데이터 분석 활성화',
enableDesc: '검색된 청크가 CSV/Excel 파일에서 온 경우, 답변 전에 LLM에 DuckDB SQL 쿼리를 생성하도록 요청합니다. LLM 호출이 한 번 더 발생하고 수 초의 지연이 추가되므로 SQL 스타일 분석이 실제로 필요할 때만 활성화하세요.',
},
fallback: {
fixed: '고정 응답',
model: '모델 생성',
+4
View File
@@ -3679,6 +3679,10 @@ export default {
boostLabel: 'Коэффициент FAQ',
boostDesc: 'Умножение оценки релевантности FAQ на этот коэффициент для повышения ранга'
},
dataAnalysis: {
enableLabel: 'Включить анализ табличных данных',
enableDesc: 'Если извлечённые фрагменты относятся к файлу CSV/Excel, перед ответом LLM сгенерирует SQL-запрос DuckDB. Это добавляет ещё один вызов LLM и несколько секунд задержки, поэтому включайте только при действительной необходимости SQL-анализа.'
},
fallback: {
fixed: 'Фиксированный ответ',
model: 'Генерация моделью'
+4
View File
@@ -4003,6 +4003,10 @@ export default {
boostLabel: "FAQ 分数加权",
boostDesc: "FAQ 结果的相关性分数乘以此系数,使其排序更靠前",
},
dataAnalysis: {
enableLabel: "启用表格数据分析",
enableDesc: "命中 CSV/Excel 类文件时,先调用大模型生成 DuckDB SQL 进行统计或筛选,再据此回答。会额外增加一次模型调用与几秒延迟,仅在确实需要时开启。",
},
fallback: {
fixed: "固定回复",
model: "模型生成",
@@ -1064,6 +1064,17 @@
</div>
</div>
</div>
<!-- 数据分析阶段开关(quick-answer 模式下,针对 CSV/Excel 文件触发额外 SQL 生成) -->
<div v-if="!isAgentMode && hasKnowledgeBase" class="setting-row">
<div class="setting-info">
<label>{{ $t('agentEditor.dataAnalysis.enableLabel') }}</label>
<p class="desc">{{ $t('agentEditor.dataAnalysis.enableDesc') }}</p>
</div>
<div class="setting-control">
<t-switch v-model="formData.config.data_analysis_enabled" />
</div>
</div>
</div>
</div>
@@ -1850,6 +1861,8 @@ const defaultFormData = {
image_storage_provider: '',
// 文件类型限制
supported_file_types: [] as string[],
// 数据分析阶段开关(默认关闭,避免在普通问答上多一次 LLM 调用生成 SQL)
data_analysis_enabled: false,
// FAQ 策略设置
faq_priority_enabled: true, // 是否启用 FAQ 优先策略
faq_direct_answer_threshold: 0.9, // FAQ 直接回答阈值(相似度高于此值直接使用 FAQ 答案)