fix MultiGradientMachine train and infer #2595
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
|
Just to confirm, will swig call c++'s defunctor when a python object is released? |
reyoung
requested changes
Jun 26, 2017
| } | ||
| } | ||
|
|
||
| void MultiGradientMachine::finish() { |
Member
Author
There was a problem hiding this comment.
After move thread->stop() to destructor, finish() seems has no use.
Member
Author
There was a problem hiding this comment.
Have add finish back
| } | ||
| } | ||
|
|
||
| MultiGradientMachine::~MultiGradientMachine() { |
Collaborator
There was a problem hiding this comment.
Maybe in destructor, we should check whether MultiGradientMachine is finished or not?
If MultiGradientMachine is not finish, then invoke this->finish();
Member
Author
@typhoonzero not sure, I will check it. |
reyoung
approved these changes
Jun 27, 2017
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
fix: #2534 #2565
Problem:
when training or infering with python v2 api, if trainer_count > 1 and call trainer.train or inferer.infer multiple times, the process will hang.
Reason:
when trainer_count > 1, paddle will use MultiGradientMachine and will start multiple worker threads to do the forward/backward work(thread number is trainer_count)
In v2 python API, trainer or inferer will call gradinet_machine.finish() after train/infer, this will stop the worker threads, when you call trianer.train or inferrer.infer the second time, there are no worker thread to handle the task, so it hangs there.
Fix:
Do not close worker threads when gradientMachine.finish() is called, close them in the destructor of gradientMachine.