super-gradients/src/super_gradients/recipes/imagenet_resnet50.yaml

#  ResNet50 Imagenet classification training:
#  This example trains with batch_size = 64 * 4 GPUs, total 256.
#  Training time on 4 x GeForce RTX A5000 is 15min / epoch.
#  Reach => 79.47 Top1 accuracy.
#
#  Log and tensorboard at s3://deci-pretrained-models/ResNet50_ImageNet/average_model.pth

# Instructions:
# running from the command line, set the PYTHONPATH environment variable: (Replace "YOUR_LOCAL_PATH" with the path to the downloaded repo):
#   export PYTHONPATH="YOUR_LOCAL_PATH"/super_gradients/
# Then:
#   python train_from_recipe_example/train_from_recipe.py --config-name=imagenet_resnet50


defaults:
  - training_hyperparams: imagenet_resnet50_train_params
  - dataset_params: imagenet_dataset_params
  - arch_params: resnet50_arch_params
  - checkpoint_params: default_checkpoint_params

arch_params:
  droppath_prob: 0.05

dataset_params:
  resize_size: 236
  random_erase_prob: 0
  random_erase_value: random
  train_interpolation: random
  rand_augment_config_string: rand-m7-mstd0.5
  cutmix: True
  cutmix_params:
    mixup_alpha: 0.2
    cutmix_alpha: 1.0
    label_smoothing: 0.1

dataset_interface:
  imagenet:
    dataset_params: ${dataset_params}

data_loader_num_workers: 8

model_checkpoints_location: local
load_checkpoint: False
checkpoint_params:
  load_checkpoint: ${load_checkpoint}

experiment_name: resnet50_imagenet

ckpt_root_dir:

multi_gpu:
  _target_: super_gradients.training.sg_model.MultiGPUMode
  value: 'DDP'

sg_model:
  _target_: super_gradients.SgModel
  experiment_name: ${experiment_name}
  model_checkpoints_location: ${model_checkpoints_location}
  ckpt_root_dir: ${ckpt_root_dir}
  multi_gpu: ${multi_gpu}

architecture: resnet50