config.json

{
  "audio": {
    "max_len": 1920000,
    "max_time": 120,
    "normalize": true,
    "random_sampling": true,
    "sample_rate": 16000,
    "skip_time": false
  },
  "augment": {
    "freq_mask_param": 8,
    "mixup_alpha": 2.5,
    "mixup_p": 0.5,
    "n_freq_masks": 1,
    "n_time_masks": 2,
    "time_freq_mask_p": 0.5,
    "time_mask_param": 8
  },
  "dataset": {
    "test_dataframe": "test.csv",
    "train_dataframe": "train.csv",
    "valid_dataframe": "valid.csv"
  },
  "environment": {
    "mixed_precision": true,
    "num_workers": 2,
    "seed": 42
  },
  "experiment_name": "spectttra_alpha-t=120",
  "logger": {
    "primary_metric": "f1",
    "project": "sonics"
  },
  "loss": {
    "label_smoothing": 0.02,
    "name": "BCEWithLogitsLoss"
  },
  "melspec": {
    "f_max": 8000,
    "f_min": 20,
    "hop_length": 512,
    "n_fft": 2048,
    "n_mels": 128,
    "norm": "mean_std",
    "power": 2,
    "top_db": 80,
    "win_length": 2048
  },
  "model": {
    "attn_drop_rate": 0.1,
    "embed_dim": 384,
    "f_clip": 1,
    "input_shape": [
      128,
      3744
    ],
    "mlp_ratio": 2.67,
    "name": "SpecTTTra",
    "num_heads": 6,
    "num_layers": 12,
    "pe_learnable": true,
    "pos_drop_rate": 0.1,
    "pre_norm": true,
    "proj_drop_rate": 0.0,
    "resume": null,
    "t_clip": 3,
    "use_init_weights": false
  },
  "num_classes": 1,
  "optimizer": {
    "clip_grad_norm": 5.0,
    "grad_accum_steps": 1,
    "momentum": 0.9,
    "opt": "adamw",
    "opt_betas": [
      0.9,
      0.999
    ],
    "opt_eps": 1e-08,
    "weight_decay": 0.05
  },
  "scheduler": {
    "decay_rate": 0.1,
    "lr": 0.0005,
    "lr_base": 0.001,
    "lr_base_scale": "linear",
    "lr_base_size": 256,
    "min_lr": 0.0,
    "sched": "cosine",
    "warmup_epochs": 5,
    "warmup_lr": 1e-06
  },
  "training": {
    "batch_size": 96,
    "epochs": 50
  },
  "validation": {
    "batch_size": 96
  }
}