[RAG] 全部LLM硬编码从deepseek-r1改为deepseek-v3;修复知识库问答source格式不匹配导致检索结果被丢弃

This commit is contained in:
2026-04-07 10:32:14 +08:00
parent 4e00a1b15e
commit a5110da4e8
12 changed files with 18 additions and 17 deletions

View File

@@ -116,15 +116,16 @@ def search_self_docs(
if top_k > 50:
data = docs
else:
# Milvus 已通过 expr 过滤了 source无需再按 fileNames 二次过滤
# Milvus 的 source 可能是原始文件名,而 fileNames 是 embeddingId格式不一致
data = [
DocumentWithVSId(
**{k: v for k, v in x[0].dict().items() if k != 'page_content'}, # 排除原有的 page_content
**{k: v for k, v in x[0].dict().items() if k != 'page_content'},
score=x[1],
id=x[0].metadata.get("id"),
page_content=f"【^[{index +1}]^ {x[0].page_content}" # 拼接索引和page_content
page_content=f"【^[{index +1}]^ {x[0].page_content}"
)
for index, x in enumerate(docs) # 使用enumerate来获取索引
if x[0].metadata.get("source") in flat_fileNames
for index, x in enumerate(docs)
]
else:
logger.warning(f"未找到知识库服务: {knowledge_base_name}")