Skip to content

Conversation

@chinakook
Copy link
Member

@chinakook chinakook commented Nov 17, 2020

I use a python property workaround to avoid the multiprocessing to pickle the target generators, which will cause error before, especially on Windows.
The error is associated with ForkingPickler of multiprocessing package of python and be usually like:
cannot pickle 'weakref' object
EOFError: Ran out of input
After the commit apache/mxnet#18328 , some memory leak were fixed. The commit is critical so we need to merge it again in MXNet 1.7.x and 1.8.x.
The revert commits below can be committed again after this fix. We can do gluon-cv training correctly and safety later.
apache/mxnet#18692
apache/mxnet@0496690

After this commit apache/mxnet#18328 , some memory leak were fixed.
Whitout this commit faster rcnn traning cannot be successfully closed.
These commits can be commit again after this yolo training fix.
apache/mxnet#18692
apache/mxnet@0496690
chinakook referenced this pull request in apache/mxnet Nov 17, 2020
@chinakook chinakook changed the title Fix yolo to support a memory leak fix Fix all generator to support a memory leak fix Nov 20, 2020
@zhreshold
Copy link
Member

@chinakook looks good to me, can you take a look at the lint error?
For self._target_generator is not callable (not-callable), it might not be the case actually, so you may disable with #pylint: disable=not-callable

@mli
Copy link
Member

mli commented Nov 24, 2020

@zhreshold zhreshold merged commit 8d5bdbb into dmlc:master Nov 24, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants