Eval Results

              precision    recall  f1-score   support

    Alakasiz       0.87      0.91      0.89       734
     Barinma       0.79      0.89      0.84       207
  Elektronik       0.69      0.83      0.75       130
       Giysi       0.71      0.81      0.76        94
    Kurtarma       0.82      0.85      0.83       362
    Lojistik       0.57      0.67      0.62       112
      Saglik       0.68      0.85      0.75       108
          Su       0.56      0.76      0.64        78
       Yagma       0.60      0.77      0.68        31
       Yemek       0.71      0.89      0.79       117

   micro avg       0.77      0.86      0.81      1973
   macro avg       0.70      0.82      0.76      1973
weighted avg       0.78      0.86      0.82      1973
 samples avg       0.83      0.88      0.84      1973

Training Params:

{'per_device_train_batch_size': 32,
 'per_device_eval_batch_size': 32,
 'learning_rate': 5.8679699888213376e-05,
 'weight_decay': 0.03530961718117487,
 'num_train_epochs': 4,
 'lr_scheduler_type': 'cosine',
 'warmup_steps': 40,
 'seed': 42,
 'fp16': True,
 'load_best_model_at_end': True,
 'metric_for_best_model': 'macro f1',
 'greater_is_better': True
}

Threshold:

  • Best Threshold: 0.40

Class Loss Weights

  • Same as Anıl's approach:
    [1.0,
     1.5167249178108022,
     1.7547338578655642,
     1.9610520059358458,
     1.8684086209021484,
     1.8019018017117145,
     2.110648663094536,
     3.081208739200435,
     1.7994815143101963]
Downloads last month
24
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Evaluation results