training_config.json

{
  "model_name": "Rajan/NepaliBERT",
  "training_strategy": "discriminative",
  "hyperparameters": {
    "batch_size": 32,
    "gradient_accumulation_steps": 1,
    "learning_rate": 3e-05,
    "num_epochs": 6,
    "max_length": 256,
    "dropout": 0.3,
    "weight_decay": 0.01,
    "label_smoothing": 0.05,
    "warmup_ratio": 0.1
  },
  "class_weights": [
    0.5,
    0.7944444444444444,
    3.0,
    3.0
  ],
  "label_names": [
    "NO",
    "OO",
    "OR",
    "OS"
  ],
  "test_metrics": {
    "model": "NepaliBERT",
    "version": "improved_v2",
    "preprocessing": "Translation + Emoji to Devanagari",
    "strategy": "discriminative",
    "hyperparameters": {
      "batch_size": 32,
      "learning_rate": 3e-05,
      "epochs": 6,
      "dropout": 0.3,
      "weight_decay": 0.01,
      "label_smoothing": 0.05,
      "warmup_ratio": 0.1
    },
    "test_metrics": {
      "accuracy": 0.6972413793103448,
      "precision": 0.493269056413594,
      "recall": 0.5412612321975574,
      "f1": 0.5125857580723172
    },
    "per_class_f1": {
      "NO": 0.7752808988764045,
      "OO": 0.6012526096033403,
      "OR": 0.48333333333333334,
      "OS": 0.19047619047619047
    },
    "overfitting_gap": 0.0659809267044067
  }
}