theDocWho's picture
Refactor: extract utils + base pipeline, expand docs
76955d6
Raw
History Blame Contribute Delete
663 Bytes
"""Small, single-purpose helpers used across `ccdp`.
Modules:
device — pick the best torch device (MPS / CUDA / CPU), seed RNGs.
transforms — torchvision transforms shared by trainers and inference.
These exist to keep the top-level training / inference code short and let any
single piece of behaviour have *one* canonical implementation.
"""
from .device import pick_device, seed_everything
from .transforms import (
IMAGENET_MEAN,
IMAGENET_STD,
eval_transform,
train_transform,
)
__all__ = [
"IMAGENET_MEAN",
"IMAGENET_STD",
"eval_transform",
"pick_device",
"seed_everything",
"train_transform",
]