imagenet

EasyTorch Example - ImageNet Classification

Reference from https://github.com/pytorch/examples/tree/main/imagenet

Train

Resnet50 (One node)

easytrain -c configs/resnet50_8x_cfg.py

MobileNet V3 Large (One node)

easytrain -c configs/mobilenet_v3_large_8x_cfg.py

Resnet50 (Muti node)

Modify CFG.DIST_INIT_METHOD='tcp://{ip_of_node_0}:{free_port}' in configs/resnet50_16x_cfg.py.

e.g.

CFG.DIST_INIT_METHOD='tcp://192.168.1.2:55555'

• Node 0:

easytrain -c configs/resnet50_16x_cfg.py

• Node 1:

easytrain -c configs/resnet50_16x_cfg.py --node-rank 1

Other models

To train other models or modify hyperparameters, customize config yourself.

Validate

Resnet50

# last
python validate.py -c configs/resnet50_8x_cfg.py --devices 0

# best
python validate.py -c configs/resnet50_8x_cfg.py --devices 0 --ckpt /path/to/ckpt_dir/resnet50_best_val_acc@1.pt

MobileNet V3 Large

python validate.py -c configs/mobilenet_v3_large_8x_cfg.py --devices 0