| "{'__module__': '__main__', 'TOKENIZED_DATA_DIR': '/kaggle/working/Hishab_Dataset/custom-pr-bn-v1-tokenized', 'OUTPUT_DIR': '/kaggle/working/models/custom-pr-bn-v1', 'LOGGING_DIR': '/kaggle/working/logs/custom-pr-bn-v1', 'MODEL_NAME': 'csebuetnlp/banglabert', 'NUM_LABELS': 8, 'LEARNING_RATE': 2e-05, 'WEIGHT_DECAY': 0.01, 'NUM_EPOCHS': 50, 'WARMUP_RATIO': 0.1, 'TRAIN_BATCH_SIZE': 16, 'EVAL_BATCH_SIZE': 32, 'GRADIENT_ACCUMULATION_STEPS': 2, 'OPTIMIZER': 'adamw_torch', 'LR_SCHEDULER': 'cosine', 'MAX_GRAD_NORM': 1.0, 'DROPOUT': 0.1, 'ATTENTION_DROPOUT': 0.1, 'HIDDEN_DROPOUT': 0.1, 'EARLY_STOPPING_PATIENCE': 5, 'EARLY_STOPPING_THRESHOLD': 0.001, 'EVAL_STRATEGY': 'epoch', 'SAVE_STRATEGY': 'epoch', 'SAVE_TOTAL_LIMIT': 3, 'LOAD_BEST_MODEL': True, 'METRIC_FOR_BEST_MODEL': 'eval_weighted_f1', 'GREATER_IS_BETTER': True, 'LOGGING_STEPS': 50, 'REPORT_TO': 'tensorboard', 'USE_CLASS_WEIGHTS': True, 'FOCAL_LOSS': True, 'FOCAL_ALPHA': 0.25, 'FOCAL_GAMMA': 2.0, 'SEED': 42, '__dict__': <attribute '__dict__' of 'TrainingConfig' objects>, '__weakref__': <attribute '__weakref__' of 'TrainingConfig' objects>, '__doc__': None}" |