pengjunjie/ragflow
1
0
Fork 1
mirror of https://github.com/infiniflow/ragflow.git synced 2025-12-08 20:42:30 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix mem leak for local reranker (#1295)

Browse Source 
### What problem does this PR solve?

#1288
### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)
This commit is contained in:
KevinHuSh
2024-06-27 14:57:24 +08:00
committed by GitHub
parent 47926a95ae
commit 0ce720a247
1 changed files with 15 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
rag/llm/rerank_model.py
View File

@ -39,6 +39,7 @@ class Base(ABC):
class DefaultRerank(Base): class DefaultRerank(Base):
_model = None _model = None
_model_lock = threading.Lock() _model_lock = threading.Lock()
def __init__(self, key, model_name, **kwargs): def __init__(self, key, model_name, **kwargs):
""" """
If you have trouble downloading HuggingFace models, -_^ this might help!! If you have trouble downloading HuggingFace models, -_^ this might help!!
@ -102,9 +103,12 @@ class JinaRerank(Base):
class YoudaoRerank(DefaultRerank): class YoudaoRerank(DefaultRerank):
_model = None _model = None
_model_lock = threading.Lock()
def __init__(self, key=None, model_name="maidalun1020/bce-reranker-base_v1", **kwargs): def __init__(self, key=None, model_name="maidalun1020/bce-reranker-base_v1", **kwargs):
from BCEmbedding import RerankerModel from BCEmbedding import RerankerModel
if not YoudaoRerank._model:
with YoudaoRerank._model_lock:
if not YoudaoRerank._model: if not YoudaoRerank._model:
try: try:
print("LOADING BCE...") print("LOADING BCE...")
@ -116,6 +120,8 @@ class YoudaoRerank(DefaultRerank):
model_name_or_path=model_name.replace( model_name_or_path=model_name.replace(
"maidalun1020", "InfiniFlow")) "maidalun1020", "InfiniFlow"))
self._model = YoudaoRerank._model
def similarity(self, query: str, texts: list): def similarity(self, query: str, texts: list):
pairs = [(query, truncate(t, self._model.max_length)) for t in texts] pairs = [(query, truncate(t, self._model.max_length)) for t in texts]
token_count = 0 token_count = 0