96 lines
3.8 KiB
Python
96 lines
3.8 KiB
Python
import os
|
|
from typing import List
|
|
import faiss
|
|
from llama_index import (
|
|
SimpleDirectoryReader,
|
|
VectorStoreIndex,
|
|
ServiceContext,
|
|
PromptTemplate,
|
|
StorageContext,
|
|
)
|
|
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
|
|
from llama_index.vector_stores.faiss import FaissVectorStore
|
|
from app.core.config import settings
|
|
from scripts.permissions import get_user_allowed_indexes
|
|
|
|
USER_INDEX_PATH = "index_data"
|
|
USER_DOC_PATH = "docs"
|
|
|
|
# ✅ 自动加前缀的 bge-m3 embedding
|
|
class BGEEmbedding(HuggingFaceEmbedding):
|
|
def _get_query_embedding(self, query: str) -> List[float]:
|
|
prefix = "Represent this sentence for searching relevant passages: "
|
|
return super()._get_query_embedding(prefix + query)
|
|
|
|
def _get_query_embeddings(self, queries: List[str]) -> List[List[float]]:
|
|
prefix = "Represent this sentence for searching relevant passages: "
|
|
return super()._get_query_embeddings([prefix + q for q in queries])
|
|
|
|
def build_user_index(user_id: str):
|
|
doc_dir = os.path.join(USER_DOC_PATH, user_id)
|
|
if not os.path.exists(doc_dir):
|
|
raise FileNotFoundError(f"文档目录不存在: {doc_dir}")
|
|
|
|
documents = SimpleDirectoryReader(doc_dir).load_data()
|
|
embed_model = BGEEmbedding(model_name=settings.MODEL_NAME)
|
|
service_context = ServiceContext.from_defaults(embed_model=embed_model, llm=None)
|
|
|
|
faiss_index = faiss.IndexFlatL2(1024)
|
|
vector_store = FaissVectorStore(faiss_index=faiss_index)
|
|
persist_dir = os.path.join(USER_INDEX_PATH, user_id)
|
|
storage_context = StorageContext.from_defaults(
|
|
persist_dir=persist_dir,
|
|
vector_store=vector_store
|
|
)
|
|
|
|
index = VectorStoreIndex.from_documents(
|
|
documents,
|
|
storage_context=storage_context,
|
|
service_context=service_context
|
|
)
|
|
index.persist(persist_dir=persist_dir)
|
|
print(f"[BUILD] 为用户 {user_id} 构建并保存了完整索引 → {persist_dir}")
|
|
|
|
def query_user_rag(user_id: str, question: str, top_k: int = 4) -> str:
|
|
embed_model = BGEEmbedding(model_name=settings.MODEL_NAME)
|
|
service_context = ServiceContext.from_defaults(embed_model=embed_model, llm=None)
|
|
|
|
persist_dir = os.path.join(USER_INDEX_PATH, user_id)
|
|
if not os.path.exists(persist_dir):
|
|
raise FileNotFoundError(f"[ERROR] 用户 {user_id} 的索引目录不存在")
|
|
|
|
storage_context = StorageContext.from_defaults(persist_dir=persist_dir)
|
|
index = VectorStoreIndex.load_from_storage(storage_context, service_context=service_context)
|
|
|
|
all_nodes = index.as_retriever(similarity_top_k=top_k).retrieve(question)
|
|
|
|
shared_indexes = get_user_allowed_indexes(user_id)
|
|
if shared_indexes:
|
|
for shared_name in shared_indexes:
|
|
shared_path = os.path.join(USER_INDEX_PATH, shared_name)
|
|
if os.path.exists(shared_path) and shared_path != persist_dir:
|
|
shared_storage = StorageContext.from_defaults(persist_dir=shared_path)
|
|
shared_index = VectorStoreIndex.load_from_storage(shared_storage, service_context=service_context)
|
|
all_nodes += shared_index.as_retriever(similarity_top_k=top_k).retrieve(question)
|
|
else:
|
|
print(f"[INFO] 用户 {user_id} 没有共享索引权限")
|
|
|
|
sorted_nodes = sorted(all_nodes, key=lambda n: -(n.score or 0))
|
|
top_nodes = sorted_nodes[:top_k]
|
|
|
|
context_str = "\n\n".join([n.get_text() for n in top_nodes])
|
|
prompt_template = PromptTemplate(
|
|
"请根据以下内容回答用户问题：\n\n{context}\n\n问题：{query}"
|
|
)
|
|
final_prompt = prompt_template.format(context=context_str, query=question)
|
|
|
|
print("[PROMPT构建完成]")
|
|
return final_prompt
|
|
|
|
if __name__ == "__main__":
|
|
uid = "user_001"
|
|
build_user_index(uid)
|
|
prompt = query_user_rag(uid, "这份资料中提到了哪些关键点？")
|
|
print("\n------ 最终构建的 Prompt 给 LLM 使用 ------\n")
|
|
print(prompt)
|