Merge pull request #109 from liangosc/ndarray-should-be-array

change the use of np.ndarray to np.array

Merge remote-tracking branch 'origin/develop' into develop

Add device_id parameter to init_process_group for CUDA device configuration

Rename wait_for_the_master to master_node_first in distributed.py

Add contextmanager import to distributed.py

Add keepdim parameter to geometric_mean and harmonic_mean functions

master_node_first

Fix wrong argument name

Fix flake8

Remove activation replacement

Bump black & reformat

Bump version

Replace deprecated torch.cuda.amp.autocast with torch.amp.autocast

vstack_header now can wrap text into multiple lines