Python 2.7 pytorch signal 137 killed "Killed" memory leak - DeepTalk - Deep Learning Community
Python 2.7 pytorch signal 137 killed “Killed” memory leak
post by sabalaba on Nov 12, 2018
Let’s say you’re running a pytorch script that seems to be crashing like this:
$ python test-pytorch.py
Killed
$ echo $?
137
Exit code 137 means that the process was killed with a signal value of 9 or SIGKILL. (137 - 128 = 9). That means your process was likely hogging system resources and was killed. The most likely situation is that you’re allocating too much memory / there’s a memory leak. I observed with htop that the memory allocation kept climbing up to the total system memory before the process was killed.
post by sabalaba on Nov 12, 2018
Oddly enough switching to python 3 solved this problem. Maybe there’s a memory leak somewhere in the py2.7 version of pytorch.
python3 test-pytorch.py
runs perfectly fine.