重构:清理废弃代码 + 优化 Agent 架构
All checks were successful
构建并部署 AI Agent 服务 / deploy (push) Successful in 5m24s
All checks were successful
构建并部署 AI Agent 服务 / deploy (push) Successful in 5m24s
主要变更: - 删除 deprecated 文件夹(intent/hybrid_router/rag_nodes 等) - 删除 intent_classifier.py(未使用) - 删除 subgraph_wrapper.py(死代码) - 重构 agent.py:简化工厂函数,支持动态模型切换 - 重构 prompts.py:添加信息获取优先级、思维链要求、工具调用约束 - 优化 tools:统一位置,rag_search 返回置信度评估 - 新增 RAG 置信度评估:embedding(25%) + rerank(25%) + LLM(50%) - 添加循环检测:防止工具无限重复调用 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
@@ -1,21 +1,33 @@
|
||||
"""
|
||||
RAG 检索流水线
|
||||
流程: 检索子文档 → 重排 → 获取父文档 → 返回
|
||||
流程: 检索子文档 → 重排 → 获取父文档 → 置信度评估 → 返回
|
||||
"""
|
||||
|
||||
import asyncio
|
||||
import re
|
||||
from dataclasses import dataclass
|
||||
from typing import List
|
||||
from langchain_core.documents import Document
|
||||
from langchain_core.language_models import BaseLanguageModel
|
||||
|
||||
from backend.app.logger import info, warning
|
||||
from ..model_services import get_rerank_service, get_small_llm_service
|
||||
from ..model_services import get_small_llm_service
|
||||
from ..rag.rerank import create_document_reranker
|
||||
from ..rag.query_transform import MultiQueryGenerator
|
||||
from ..rag.fusion import reciprocal_rank_fusion
|
||||
from ..rag.retriever import create_parent_hybrid_retriever
|
||||
|
||||
|
||||
@dataclass
|
||||
class RAGResult:
|
||||
"""RAG 检索结果(包含置信度)"""
|
||||
content: str # 格式化后的上下文
|
||||
documents: List[Document] # 原始文档
|
||||
confidence: float # 综合置信度 0.0-1.0
|
||||
scores: dict # 各维度分数 {embedding, rerank, llm, final}
|
||||
is_useful: bool # 是否可用(confidence >= 0.6)
|
||||
|
||||
|
||||
class RAGPipeline:
|
||||
def __init__(
|
||||
self,
|
||||
@@ -26,6 +38,7 @@ class RAGPipeline:
|
||||
collection_name: str = "rag_documents",
|
||||
use_rerank: bool = True,
|
||||
return_parent_docs: bool = True,
|
||||
confidence_threshold: float = 0.6,
|
||||
):
|
||||
self.retriever = retriever or create_parent_hybrid_retriever(
|
||||
collection_name=collection_name, search_k=rerank_top_n * 4
|
||||
@@ -34,8 +47,9 @@ class RAGPipeline:
|
||||
self.rerank_top_n = rerank_top_n
|
||||
self.use_rerank = use_rerank
|
||||
self.return_parent_docs = return_parent_docs
|
||||
self._last_docs = [] # 保存最后一次检索的文档
|
||||
self._last_scores = [] # 保存最后一次检索的分数
|
||||
self.confidence_threshold = confidence_threshold
|
||||
self._last_docs: List[Document] = []
|
||||
self._last_scores: List[dict] = []
|
||||
|
||||
if llm == "default_small":
|
||||
try:
|
||||
@@ -47,62 +61,188 @@ class RAGPipeline:
|
||||
|
||||
self.query_generator = MultiQueryGenerator(self.llm, num_queries) if self.llm else None
|
||||
self.reranker = create_document_reranker() if use_rerank else None
|
||||
info(f"[Pipeline] init: rerank={use_rerank}, return_parent={return_parent_docs}")
|
||||
info(f"[Pipeline] init: rerank={use_rerank}, return_parent={return_parent_docs}, threshold={confidence_threshold}")
|
||||
|
||||
@property
|
||||
def last_docs(self) -> List[Document]:
|
||||
"""获取最后一次检索的文档"""
|
||||
return self._last_docs
|
||||
|
||||
@property
|
||||
def last_scores(self) -> List[dict]:
|
||||
"""获取最后一次检索的分数信息"""
|
||||
return self._last_scores
|
||||
|
||||
async def aretrieve(self, query: str) -> List[Document]:
|
||||
info(f"[Pipeline] aretrieve 开始: query={query[:50]}...")
|
||||
# Step 1: 检索
|
||||
info(f"[Pipeline] Step 1: 调用 _retrieve")
|
||||
child_docs = await self._retrieve(query)
|
||||
info(f"[Pipeline] Step 1 完成: 检索到 {len(child_docs)} 个子文档")
|
||||
# 调试:打印子文档长度
|
||||
for i, doc in enumerate(child_docs[:5]):
|
||||
content_len = len(doc.page_content)
|
||||
info(f"[Pipeline] 子文档[{i}] 长度={content_len}字符")
|
||||
"""原接口,保持向后兼容"""
|
||||
docs = await self._do_retrieve(query)
|
||||
self._last_docs = docs
|
||||
self._last_scores = self._extract_scores(docs)
|
||||
return docs
|
||||
|
||||
# Step 1.5: 向量初筛(进入重排前先过滤)
|
||||
async def aretrieve_with_confidence(self, query: str, original_query: str = "") -> RAGResult:
|
||||
"""
|
||||
带置信度评估的检索
|
||||
|
||||
Args:
|
||||
query: 检索查询
|
||||
original_query: 原始用户问题(用于置信度评估)
|
||||
|
||||
Returns:
|
||||
RAGResult: 包含内容和置信度的结构化结果
|
||||
"""
|
||||
info(f"[Pipeline] aretrieve_with_confidence: query={query[:50]}...")
|
||||
|
||||
# 1. 执行检索
|
||||
docs = await self._do_retrieve(query)
|
||||
self._last_docs = docs
|
||||
self._last_scores = self._extract_scores(docs)
|
||||
|
||||
# 2. 格式化内容
|
||||
content = self.format_context(docs)
|
||||
|
||||
if not docs or not content:
|
||||
info(f"[Pipeline] 无检索结果,置信度=0")
|
||||
return RAGResult(
|
||||
content="",
|
||||
documents=[],
|
||||
confidence=0.0,
|
||||
scores={"embedding": 0.0, "rerank": 0.0, "llm": 0.0, "final": 0.0},
|
||||
is_useful=False
|
||||
)
|
||||
|
||||
# 3. 评估置信度(三维度)
|
||||
scores = await self._evaluate_confidence(
|
||||
query=original_query or query,
|
||||
docs=docs,
|
||||
content=content
|
||||
)
|
||||
|
||||
confidence = scores["final"]
|
||||
is_useful = confidence >= self.confidence_threshold
|
||||
|
||||
info(f"[Pipeline] 置信度评估完成: confidence={confidence:.3f}, is_useful={is_useful}")
|
||||
|
||||
return RAGResult(
|
||||
content=content,
|
||||
documents=docs,
|
||||
confidence=confidence,
|
||||
scores=scores,
|
||||
is_useful=is_useful
|
||||
)
|
||||
|
||||
async def _do_retrieve(self, query: str) -> List[Document]:
|
||||
"""执行检索流程"""
|
||||
# Step 1: 检索
|
||||
child_docs = await self._retrieve(query)
|
||||
|
||||
# Step 1.5: 向量初筛
|
||||
vector_top_n = 20
|
||||
info(f"[Pipeline] Step 1.5: 向量初筛: 取前 {vector_top_n} 个(当前 {len(child_docs)} 个)")
|
||||
if len(child_docs) > vector_top_n:
|
||||
child_docs = child_docs[:vector_top_n]
|
||||
info(f"[Pipeline] Step 1.5 完成: 向量初筛后 {len(child_docs)} 个")
|
||||
|
||||
# Step 2: 重排
|
||||
info(f"[Pipeline] Step 2: 开始重排")
|
||||
if self.reranker:
|
||||
try:
|
||||
child_docs = self.reranker.compress_documents(child_docs, query, self.rerank_top_n)
|
||||
info(f"[Pipeline] Step 2 完成: 重排后 {len(child_docs)} 个")
|
||||
except Exception as e:
|
||||
warning(f"[Pipeline] 重排失败: {e}")
|
||||
child_docs = child_docs[:self.rerank_top_n]
|
||||
else:
|
||||
info(f"[Pipeline] Step 2 跳过: 未启用 reranker")
|
||||
|
||||
# Step 3: 获取父文档
|
||||
info(f"[Pipeline] Step 3: 开始获取父文档")
|
||||
if self.return_parent_docs:
|
||||
parent_docs = await self._get_parents(child_docs)
|
||||
info(f"[Pipeline] Step 3 完成: 获取到 {len(parent_docs)} 个父文档")
|
||||
# 保存分数信息到 last_scores 供外部访问
|
||||
self._last_scores = self._extract_scores(parent_docs)
|
||||
info(f"[Pipeline] aretrieve 结束: 返回父文档")
|
||||
return parent_docs
|
||||
|
||||
self._last_scores = self._extract_scores(child_docs)
|
||||
info(f"[Pipeline] aretrieve 结束: 返回子文档")
|
||||
return await self._get_parents(child_docs)
|
||||
return child_docs
|
||||
|
||||
async def _evaluate_confidence(
|
||||
self,
|
||||
query: str,
|
||||
docs: List[Document],
|
||||
content: str
|
||||
) -> dict:
|
||||
"""
|
||||
三维度置信度评估
|
||||
|
||||
Returns:
|
||||
{
|
||||
"embedding": float, # 向量相似度 (0-1)
|
||||
"rerank": float, # 重排分数 (0-1)
|
||||
"llm": float, # LLM判断 (0-1)
|
||||
"final": float # 综合分数 (0-1)
|
||||
}
|
||||
"""
|
||||
scores = {"embedding": 0.0, "rerank": 0.0, "llm": 0.5, "final": 0.0}
|
||||
|
||||
# 1. 向量相似度
|
||||
if docs:
|
||||
embedding_scores = []
|
||||
for doc in docs:
|
||||
score = doc.metadata.get("embedding_score", doc.metadata.get("score", 0.0))
|
||||
# 归一化(如果分数 > 1)
|
||||
if score > 1.0:
|
||||
score = min(score / 10.0, 1.0)
|
||||
embedding_scores.append(score)
|
||||
scores["embedding"] = max(embedding_scores) if embedding_scores else 0.0
|
||||
|
||||
info(f"[Confidence] embedding={scores['embedding']:.3f}")
|
||||
|
||||
# 2. 重排分数
|
||||
if docs:
|
||||
rerank_scores = []
|
||||
for doc in docs:
|
||||
score = doc.metadata.get("rerank_score", 0.0)
|
||||
# 归一化(假设满分 10)
|
||||
if score > 1.0:
|
||||
score = min(score / 10.0, 1.0)
|
||||
rerank_scores.append(score)
|
||||
scores["rerank"] = max(rerank_scores) if rerank_scores else 0.0
|
||||
|
||||
info(f"[Confidence] rerank={scores['rerank']:.3f}")
|
||||
|
||||
# 3. LLM 判断
|
||||
if self.llm and content:
|
||||
llm_score = await self._get_llm_confidence(query, content)
|
||||
scores["llm"] = llm_score
|
||||
info(f"[Confidence] llm={scores['llm']:.3f}")
|
||||
|
||||
# 4. 综合得分(加权平均)
|
||||
scores["final"] = (
|
||||
scores["embedding"] * 0.25 +
|
||||
scores["rerank"] * 0.25 +
|
||||
scores["llm"] * 0.50
|
||||
)
|
||||
|
||||
info(f"[Confidence] final={scores['final']:.3f}")
|
||||
|
||||
return scores
|
||||
|
||||
async def _get_llm_confidence(self, query: str, context: str) -> float:
|
||||
"""使用 LLM 评估检索结果相关性"""
|
||||
try:
|
||||
prompt = f"""评估以下检索结果与用户问题的相关性,返回 0.0-1.0 的分数:
|
||||
- 1.0 = 完全相关,能直接回答问题
|
||||
- 0.7 = 高度相关,有很大参考价值
|
||||
- 0.5 = 部分相关,有一定参考价值
|
||||
- 0.3 = 低度相关,参考价值有限
|
||||
- 0.0 = 完全不相关,无法回答问题
|
||||
|
||||
用户问题:{query}
|
||||
|
||||
检索结果:{context[:1500]}
|
||||
|
||||
只返回一个数字(0.0-1.0):"""
|
||||
|
||||
response = await self.llm.ainvoke(prompt)
|
||||
content = response.content.strip()
|
||||
|
||||
match = re.search(r'(\d+\.?\d*)', content)
|
||||
if match:
|
||||
score = float(match.group(1))
|
||||
return max(0.0, min(1.0, score))
|
||||
|
||||
except Exception as e:
|
||||
info(f"[Confidence] LLM评估失败: {e}")
|
||||
|
||||
return 0.5 # 默认中等置信度
|
||||
|
||||
def _extract_scores(self, docs: List[Document]) -> List[dict]:
|
||||
"""提取文档的分数信息"""
|
||||
scores = []
|
||||
@@ -114,84 +254,54 @@ class RAGPipeline:
|
||||
return scores
|
||||
|
||||
async def _retrieve(self, query: str) -> List[Document]:
|
||||
info(f"[Pipeline] _retrieve 开始: query={query[:50]}...")
|
||||
if self.query_generator:
|
||||
info(f"[Pipeline] _retrieve: 调用 query_generator.agenerate")
|
||||
queries = await self.query_generator.agenerate(query)
|
||||
queries = [query] + [q for q in queries if q != query]
|
||||
info(f"[Pipeline] _retrieve: 生成 {len(queries)} 个查询: {queries}")
|
||||
info(f"[Pipeline] _retrieve: 开始 asyncio.gather 并行检索")
|
||||
doc_lists = await asyncio.gather(*[self.retriever.ainvoke(q) for q in queries])
|
||||
info(f"[Pipeline] _retrieve: asyncio.gather 完成,得到 {len(doc_lists)} 组结果")
|
||||
info(f"[Pipeline] _retrieve: 开始 reciprocal_rank_fusion")
|
||||
result = reciprocal_rank_fusion(doc_lists)
|
||||
info(f"[Pipeline] _retrieve: RRF 完成,得到 {len(result)} 个文档")
|
||||
info(f"[Pipeline] _retrieve 结束")
|
||||
return result
|
||||
info(f"[Pipeline] _retrieve: query_generator 未启用,直接单次检索")
|
||||
result = await self.retriever.ainvoke(query)
|
||||
info(f"[Pipeline] _retrieve 结束")
|
||||
return result
|
||||
return reciprocal_rank_fusion(doc_lists)
|
||||
return await self.retriever.ainvoke(query)
|
||||
|
||||
async def _get_parents(self, child_docs: List[Document]) -> List[Document]:
|
||||
info(f"[Pipeline] _get_parents 开始: {len(child_docs)} 个子文档")
|
||||
# 收集 parent_id 和对应的分数
|
||||
parent_map = {} # parent_id -> (embedding_score, rerank_score)
|
||||
|
||||
parent_map = {}
|
||||
for doc in child_docs:
|
||||
pid = doc.metadata.get("parent_id")
|
||||
if pid and pid not in parent_map:
|
||||
# embedding 分数
|
||||
embedding_score = doc.metadata.get("score", 0.0)
|
||||
# rerank 分数(如果有的话)
|
||||
rerank_score = doc.metadata.get("rerank_score", 0.0)
|
||||
parent_map[pid] = (embedding_score, rerank_score)
|
||||
|
||||
info(f"[Pipeline] _get_parents: 收集到 {len(parent_map)} 个 unique parent_id")
|
||||
if not parent_map:
|
||||
warning("[Pipeline] 未找到 parent_id,返回子文档")
|
||||
return child_docs
|
||||
|
||||
try:
|
||||
info(f"[Pipeline] _get_parents: 调用 create_docstore")
|
||||
from backend.rag_core import create_docstore
|
||||
docstore, _ = create_docstore()
|
||||
info(f"[Pipeline] _get_parents: 调用 docstore.amget")
|
||||
parent_docs =await docstore.amget(list(parent_map.keys()))
|
||||
info(f"[Pipeline] _get_parents: docstore.amget 返回 {len(parent_docs)} 个结果")
|
||||
parent_docs = await docstore.amget(list(parent_map.keys()))
|
||||
|
||||
# 构建结果,保持分数信息
|
||||
result = []
|
||||
for doc in parent_docs:
|
||||
if doc:
|
||||
pid = doc.metadata.get("id")
|
||||
scores = parent_map.get(pid, (0.0, 0.0))
|
||||
# 将分数添加到 metadata 中
|
||||
doc.metadata["embedding_score"] = scores[0]
|
||||
doc.metadata["rerank_score"] = scores[1]
|
||||
result.append((doc, scores[0] + scores[1] * 2)) # 综合分数,rerank 权重更高
|
||||
# 综合分数,rerank 权重更高
|
||||
result.append((doc, scores[0] + scores[1] * 2))
|
||||
|
||||
result.sort(key=lambda x: x[1], reverse=True)
|
||||
docs = [d for d, _ in result]
|
||||
info(f"[Pipeline] _get_parents: 最终得到 {len(docs)} 个父文档")
|
||||
info(f"[Pipeline] _get_parents 结束")
|
||||
return docs
|
||||
return [d for d, _ in result]
|
||||
except Exception as e:
|
||||
warning(f"[Pipeline] 获取父文档失败: {e}", exc_info=True)
|
||||
warning(f"[Pipeline] 获取父文档失败: {e}")
|
||||
return child_docs
|
||||
|
||||
def format_context(self, documents: List[Document]) -> str:
|
||||
info(f"[Pipeline] format_context 开始: {len(documents)} 个文档")
|
||||
if not documents:
|
||||
info(f"[Pipeline] format_context: 无文档,返回空字符串")
|
||||
return ""
|
||||
parts = []
|
||||
for i, doc in enumerate(documents, 1):
|
||||
source = doc.metadata.get("source", "未知来源")
|
||||
parts.append(f"【资料 {i}】来源:{source}\n{doc.page_content}\n---\n")
|
||||
result = "\n".join(parts)
|
||||
info(f"[Pipeline] format_context 结束: 结果长度={len(result)} 字符")
|
||||
return result
|
||||
return "\n".join(parts)
|
||||
|
||||
|
||||
def create_rag_pipeline(**kwargs) -> RAGPipeline:
|
||||
|
||||
Reference in New Issue
Block a user