training_args.json

{
  "base_model": "roberta-base",
  "data": {
    "csv_path": "mesh_dataset.csv",
    "split": "train=0.9, val=0.1",
    "random_state": 42
  },
  "hyperparams": {
    "batch_size": 32,
    "lr": 0.0001,
    "num_epochs_max": 10,
    "patience": 3,
    "min_delta": 0.0001,
    "lora_r": 16,
    "lora_alpha": 32,
    "lora_dropout": 0.1,
    "target_modules": [
      "query",
      "key",
      "value",
      "output.dense"
    ]
  },
  "best_val_loss": 0.008271555498198692
}