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

#  RegnetY Imagenet classification training:
#  This example trains with batch_size = 256
#  Training time on a single GeForce RTX 2080 Ti, and top1 accuracies:
#  11 days for RegnetY200, 70.88
#  12 days for RegnetY400, 74.74
#  19 days for RegnetY600, 76.18
#  20 days for RegnetY800, 77.07
#  NOTE: Training should probably be lower as resources were shared among the above runs.

#  Logs and tensorboards at:
# https://deci-pretrained-models.s3.amazonaws.com/RegnetY800/
# https://deci-pretrained-models.s3.amazonaws.com/RegnetY600/
# https://deci-pretrained-models.s3.amazonaws.com/RegnetY400/
# https://deci-pretrained-models.s3.amazonaws.com/RegnetY200/

# Instructions:
# Set the PYTHONPATH environment variable: (Replace "YOUR_LOCAL_PATH" with the path to the downloaded repo):
#   export PYTHONPATH="YOUR_LOCAL_PATH"/super_gradients/:"YOUR_LOCAL_PATH"/super_gradients/src/
# Then:
#   python train_from_recipe_example/train_from_recipe.py --config-name=imagenet_regnetY architecture: regnetY200 experiment_name: regnetY200_imagenet
#   python train_from_recipe_example/train_from_recipe.py --config-name=imagenet_regnetY architecture: regnetY400 experiment_name: regnetY400_imagenet
#   python train_from_recipe_example/train_from_recipe.py --config-name=imagenet_regnetY architecture: regnetY600 experiment_name: regnetY600_imagenet
#   python train_from_recipe_example/train_from_recipe.py --config-name=imagenet_regnetY architecture: regnetY800 experiment_name: regnetY800_imagenet

defaults:
  - training_hyperparams: imagenet_regnetY_train_params
  - dataset_params: imagenet_dataset_params
  - arch_params: regnetY_arch_params
  - checkpoint_params: default_checkpoint_params

arch_params:   
  num_classes: 1000
  dropout_prob: 0.5
  droppath_prob: 0.0

dataset_params:
  batch_size: 256
  color_jitter: 0.4
  random_erase_prob: 0.2
  random_erase_value: random
  train_interpolation: random
  auto_augment_config_string: rand-m9-mstd0.5

dataset_interface:
  _target_: super_gradients.training.datasets.dataset_interfaces.dataset_interface.ImageNetDatasetInterface
  dataset_params: ${dataset_params}
  data_dir: /data/Imagenet

data_loader_num_workers: 8

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

experiment_name: regnetY800_imagenet

ckpt_root_dir:

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

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: regnetY800