zamba
zamba copied to clipboard
Support `auto_lr_find` with `num_workers` > 0
auto_lr_find does not support DDP. We set the multiprocessing context based on the number of workers. This leads to a scenario where a user cannot use auto_lr_find with multiple workers, even though this would be possible if the multiprocessing context were set to None.
A couple options:
- infer the multiprocessing context based on number of GPUs instead of number of workers
- allow the user to explicitly set the multiprocessing context
- do not support multiprocessing until PTL supports this for auto lr find