nsynth_clf.yaml

pretrain_experiment_dir: null
downstream_experiment_id: null
downstream_dir: ${pretrain_experiment_dir}/downstream
dataset_name: nsynth
task: classification
backbone_init: mulap
freeze_backbone: true
feature_dim: null   # dimension of audio embeddings when backbone_init = "no_backbone"
bert_audio_embed: false
classifier: mlp
hidden_dim: 512 # dimension of hidden layer in classifier if using MLP
dropout_p: 0.0

training:
  # Maximum number of epochs
  epochs: 200
  # Device on which the model will be trained. Set 'cpu' to train/infer on CPU
  device: cuda
  # Configuration for data loading 
  # Size of each batch
  batch_size: 64
  # Number of workers to be used in dataloader
  num_workers: 4
  # Whether to shuffle in dataloader
  shuffle: true
  # Whether to pin memory in dataloader
  pin_memory: true
  # Configuration for optimizer
  optimizer: adam
  # Weight decay for Adam
  weight_decay: 0.001
  # Initial learning rate
  lr: 0.001
  # Early stopping configs
  early_stop:
    enabled: true
    criteria: task_metric
    minimize: false
    patience: 50