|
|
|
@ -5,6 +5,7 @@ import uuid
|
|
|
|
from contextlib import asynccontextmanager
|
|
|
|
from contextlib import asynccontextmanager
|
|
|
|
from io import BytesIO
|
|
|
|
from io import BytesIO
|
|
|
|
from logging.handlers import RotatingFileHandler
|
|
|
|
from logging.handlers import RotatingFileHandler
|
|
|
|
|
|
|
|
from typing import List
|
|
|
|
|
|
|
|
|
|
|
|
import jieba # 导入 jieba 分词库
|
|
|
|
import jieba # 导入 jieba 分词库
|
|
|
|
import uvicorn
|
|
|
|
import uvicorn
|
|
|
|
@ -83,7 +84,7 @@ def text_to_embedding(text):
|
|
|
|
return [0.0] * model.vector_size
|
|
|
|
return [0.0] * model.vector_size
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
async def generate_stream(client, milvus_pool, collection_manager, query):
|
|
|
|
async def generate_stream(client, milvus_pool, collection_manager, query, documents):
|
|
|
|
# 从连接池获取连接
|
|
|
|
# 从连接池获取连接
|
|
|
|
connection = milvus_pool.get_connection()
|
|
|
|
connection = milvus_pool.get_connection()
|
|
|
|
try:
|
|
|
|
try:
|
|
|
|
@ -95,8 +96,17 @@ async def generate_stream(client, milvus_pool, collection_manager, query):
|
|
|
|
"metric_type": "L2", # 使用 L2 距离度量方式
|
|
|
|
"metric_type": "L2", # 使用 L2 距离度量方式
|
|
|
|
"params": {"nprobe": MS_NPROBE} # 设置 IVF_FLAT 的 nprobe 参数
|
|
|
|
"params": {"nprobe": MS_NPROBE} # 设置 IVF_FLAT 的 nprobe 参数
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
# 动态生成expr表达式
|
|
|
|
|
|
|
|
if documents:
|
|
|
|
|
|
|
|
conditions = [f"array_contains(tags['tags'], '{doc}')" for doc in documents]
|
|
|
|
|
|
|
|
expr = " OR ".join(conditions)
|
|
|
|
|
|
|
|
else:
|
|
|
|
|
|
|
|
expr = "" # 如果没有选择文档，返回空字符串
|
|
|
|
# 7. 将文本转换为嵌入向量
|
|
|
|
# 7. 将文本转换为嵌入向量
|
|
|
|
results = collection_manager.search(current_embedding, search_params, limit=5) # 返回 5 条结果
|
|
|
|
results = collection_manager.search(current_embedding,
|
|
|
|
|
|
|
|
search_params,
|
|
|
|
|
|
|
|
expr=expr, # 使用in操作符
|
|
|
|
|
|
|
|
limit=5) # 返回 5 条结果
|
|
|
|
|
|
|
|
|
|
|
|
# 3. 处理搜索结果
|
|
|
|
# 3. 处理搜索结果
|
|
|
|
logger.info("最相关的知识库内容:")
|
|
|
|
logger.info("最相关的知识库内容:")
|
|
|
|
@ -176,6 +186,7 @@ http://10.10.21.22:8000/static/ai.html
|
|
|
|
|
|
|
|
|
|
|
|
class QueryRequest(BaseModel):
|
|
|
|
class QueryRequest(BaseModel):
|
|
|
|
query: str = Field(..., description="用户查询的问题")
|
|
|
|
query: str = Field(..., description="用户查询的问题")
|
|
|
|
|
|
|
|
documents: List[str] = Field(..., description="用户上传的文档")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class SaveWordRequest(BaseModel):
|
|
|
|
class SaveWordRequest(BaseModel):
|
|
|
|
@ -249,7 +260,8 @@ async def rag_stream(request: Request):
|
|
|
|
request.app.state.deepseek_client,
|
|
|
|
request.app.state.deepseek_client,
|
|
|
|
request.app.state.milvus_pool,
|
|
|
|
request.app.state.milvus_pool,
|
|
|
|
request.app.state.collection_manager,
|
|
|
|
request.app.state.collection_manager,
|
|
|
|
query_request.query
|
|
|
|
query_request.query,
|
|
|
|
|
|
|
|
query_request.documents
|
|
|
|
):
|
|
|
|
):
|
|
|
|
return chunk
|
|
|
|
return chunk
|
|
|
|
|
|
|
|
|
|
|
|
|
