super-gradients/recipes/imagenet_repvgg.yaml

#  RepVGGA0 Imagenet classification training:
#  This example trains with batch_size = 64 * 4 GPUs, total 256.
#  Training time on 4 X GeForce RTX 3090 Ti is 10min / epoch, total time ~ 20h 22m (DistributedDataParallel).
#  Reach => 72.05 Top1 accuracy.
#
#  Log and tensorboard at s3://deci-pretrained-models/repvggg-a0-imagenet-tensorboard/

# Instructions:
# Set the PYTHONPATH environment variable: (Replace "YOUR_LOCAL_PATH" with the path to the downloaded repo):
#   export PYTHONPATH="YOUR_LOCAL_PATH"/super_gradients/
# Then for 320x320 image size for training:
#   python -m torch.distributed.launch --nproc_per_node=4 train_from_recipe_example/train_from_recipe.py --config-name=imagenet_repvgg

defaults:
  - training_hyperparams: imagenet_repvgg_train_params
  - dataset_params: imagenet_dataset_params
  - arch_params: default_arch_params

arch_params:
  build_residual_branches: True

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

experiment_name: repvgg_a0_imagenet_reproduce_fix

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

deci_model:
  _target_: super_gradients.SgModel
  experiment_name: ${experiment_name}
  multi_gpu: ${multi_gpu}

architecture: repvgg_a0