training_log.json

{
  "experiment_id": "bert_dual_head_20260213_060359",
  "config": {
    "model_name": "bert-base-uncased",
    "num_category_classes": 7,
    "num_severity_classes": 4,
    "batch_size": 16,
    "learning_rate": 0.0003,
    "num_epochs": 3,
    "max_length": 128,
    "lora_r": 8,
    "lora_alpha": 16,
    "lora_dropout": 0.1,
    "seed": 42
  },
  "device": "cuda",
  "epochs": [
    {
      "epoch": 1,
      "train_loss": 2.1082094073295594,
      "val_loss": 1.4672169902107932,
      "val_category_acc": 0.6914285714285714,
      "val_severity_acc": 0.7457142857142857
    },
    {
      "epoch": 2,
      "train_loss": 1.4187353157997131,
      "val_loss": 1.252626198259267,
      "val_category_acc": 0.7628571428571429,
      "val_severity_acc": 0.7771428571428571
    },
    {
      "epoch": 3,
      "train_loss": 1.2215006688662937,
      "val_loss": 1.1916442256082187,
      "val_category_acc": 0.7628571428571429,
      "val_severity_acc": 0.7828571428571428
    }
  ],
  "final_test_metrics": {
    "loss": 1.1840462291782552,
    "category_accuracy": 0.7628571428571429,
    "severity_accuracy": 0.7828571428571428
  }
}