`TOR401` give more context on what to set `num_workers`

[jamesbraza]

I am seeing this TOR401 error message as of TorchFix==0.7.0:

TOR401 Detected DataLoader running with synchronized implementation. Please enable asynchronous dataloading by setting num_workers > 0 when initializing DataLoader.

For reference, from https://github.com/pytorch/pytorch/blob/v2.6.0/torch/utils/data/dataloader.py#L153-L155:

num_workers (int, optional): how many subprocesses to use for data loading. 0 means that the data will be loaded in the main process.

Can you expand this TOR401 error message (or add TOR401 docs) advising what to set num_workers? Is a generally useful to set it to 1, or to cpu_count?

[jamesbraza]

Cc @kit1980 since he implemented this rule in c99ad67