sampling_model parameters update

[gcrth]

I am not sure if this is really a problem. There is a line

gflownet/src/gflownet/online_trainer.py

Line 133 in d04f9e1

b.data.mul_(self.sampling_tau).add_(a.data * (1 - self.sampling_tau))

. If there are multiple workers in the dataloader, updating parameters like this may not be thread safe since sampling_model is used during the training data generation. I assume some kind of lock may be needed.

[gcrth]

'prefetch_factor'

gflownet/src/gflownet/trainer.py

Line 150 in d04f9e1

prefetch_factor=1 if self.cfg.num_workers else None,

is also related to this issue.