training_config.json

{
  "model_name": "microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext",
  "task": "text-classification",
  "num_labels": 9,
  "classes": [
    "CASE_REPORTS",
    "COMMENT",
    "EDITORIAL",
    "GUIDELINES",
    "META_ANALYSIS",
    "PROSPECTIVE",
    "RCT",
    "RETROSPECTIVE",
    "REVIEW"
  ],
  "training_details": {
    "final_accuracy": 0.937962962962963,
    "final_f1": 0.9377,
    "total_samples": 54000,
    "epochs": 10,
    "learning_rate": 0.0001,
    "batch_size": 32,
    "max_length": 512,
    "best_checkpoint": "checkpoint-2000"
  },
  "text_preprocessing": {
    "format": "[TITLE] {title} [SEP] [ABSTRACT] {abstract}",
    "special_tokens": [
      "[TITLE]",
      "[SEP]",
      "[ABSTRACT]"
    ],
    "truncation_strategy": "optimized_title_abstract_split"
  }
}