training_config.json

{
  "model_name": "UBC-NLP/MARBERTv2",
  "data_dir": "data2/",
  "output_dir": "output/model/final",
  "max_length": 256,
  "seed": 42,
  "train_batch_size": 64,
  "eval_batch_size": 128,
  "learning_rate": 2e-05,
  "weight_decay": 0.01,
  "num_epochs": 5,
  "warmup_ratio": 0.1,
  "gradient_accumulation_steps": 1,
  "eval_strategy": "epoch",
  "save_strategy": "epoch",
  "logging_steps": 50,
  "early_stopping_patience": 2,
  "metric_for_best_model": "f1_macro",
  "greater_is_better": true,
  "fp16": true,
  "remove_html_tags": true,
  "remove_urls": false,
  "remove_mentions": false,
  "remove_tatweel": true,
  "remove_diacritics": false,
  "normalize_alef": false,
  "normalize_teh_marbuta": false,
  "normalize_yeh": false
}