cifar100_backbone_old.yaml

arch_params:
  arch: ResNet50
  num_classes: 100 # This is the number of classes for architecture FC layer.
  embedding_dim: 128
  last_nonlin: True

optimizer_params:
  algorithm: sgd
  lr: 1.024
  weight_decay: 0.000030517578125
  no_bn_decay: False
  momentum: 0.875
  nesterov: False

dataset_params:
  name: cifar100
  data_root: data_store/cifar-100-python # This should contain training and validation dirs.
  num_classes: 50 # This is the number of classes to include for training.
  num_workers: 20
  batch_size: 1024

lr_policy_params:
  algorithm: cosine_lr
  warmup_length: 5
  epochs: 100
  lr: 1.024

epochs: 100
label_smoothing: 0.1
output_model_path: checkpoints/cifar100_old.pt