进程退出 cleanup 时抛出异常

[patricksuo]

环境:

• Mac silicon
• Python 3.12.8
• FlagEmbedding==1.3.3

:Exception ignored in: <function AbsEmbedder.__del__ at 0x11bb863e0>
Traceback (most recent call last):
  File "/xxx/.venv/lib/python3.12/site-packages/FlagEmbedding/abc/inference/AbsEmbedder.py", line 270, in __del__
  File "/xxx/.venv/lib/python3.12/site-packages/FlagEmbedding/abc/inference/AbsEmbedder.py", line 89, in stop_self_pool
TypeError: 'NoneType' object is not callable

[patricksuo]

del() can be executed during interpreter shutdown. As a consequence, the global variables it needs to access (including other modules) may already have been deleted or set to None. Python guarantees that globals whose name begins with a single underscore are deleted from their module before other globals are deleted; if no other references to such globals exist, this may help in assuring that imported modules are still available at the time when the del() method is called.

FlagEmbedding/FlagEmbedding/abc/inference/AbsEmbedder.py

Lines 99 to 108 in 0306f56

	def stop_self_pool(self):
	if self.pool is not None:
	self.stop_multi_process_pool(self.pool)
	self.pool = None
	try:
	self.model.to('cpu')
	torch.cuda.empty_cache()
	except:
	pass
	gc.collect()

在进程退出时，这段 cleanup 代码会被执行，但这时候 gc.collect 已经被设置为 None 了，所以会抛出异常

[hanhainebula]

你好，@patricksuo，谢谢你提出这个问题！我们的测试中还没有遇到过这类问题，可以麻烦你给一段代码帮助我们复现这个问题吗？谢谢！

[patricksuo]

@hanhainebula
我本地是这样重现的：
1）加载模型
2）把模型放到全局变量
3) 调用 encode query
3）进程退出

from langchain_text_splitters import CharacterTextSplitter
from FlagEmbedding import BGEM3FlagModel


model = BGEM3FlagModel('BAAI/bge-m3', return_dense=True, return_sparse=True)


def search(query:str) :
    query_vector = model.encode_queries([query], batch_size=8, return_dense=True, return_sparse=True, convert_to_numpy=True)['dense_vecs']
    return query_vector
   
search("xxx")

进程退出时 gc.collect 被置为 None 了

[patricksuo]

今天再试了一下，比较奇怪，可能不是一个可以确定复现的问题。