Fix dataloader
We have an issue with the worker: some processes are blocked in "Running" status for ages.
We've made some tests, and it looks like this issue only occurs when num_workers>0
.
-
num_workers=0
- working fine -
num_workers=2
- blocked -
num_workers=4
- blocked
This is a parameter from Pytorch's dataloader:
num_workers (int, optional) – how many subprocesses to use for data loading. 0 means that the data will be loaded in the main process. (default: 0)
So the aim of this MR is to :
- set
num_workers=0
by default - remove the key
num_workers
from user configuration