---
library_name: transformers
language:
- en
base_model:
- microsoft/deberta-base
pipeline_tag: text-classification
---

# Model Card

This model is a finetuned version of [microsoft/deberta-base](https://huggingface.co/microsoft/deberta-base) on the [Onion or Not](https://www.kaggle.com/datasets/chrisfilo/onion-or-not) dataset.
The model was fine-tuned for 5 epochs with a learning rate of 2e-5 and a linear schedule. Random token dropout was implemented during training to avoid overfitting.
The classification report is shown below:
```
Final Validation Accuracy: 93.31%

Final Classification Report:
              precision    recall  f1-score   support

    NotOnion       0.94      0.96      0.95      3000
       Onion       0.93      0.89      0.91      1800

    accuracy                           0.93      4800
   macro avg       0.93      0.92      0.93      4800
weighted avg       0.93      0.93      0.93      4800
```

Running inference on a new sample gave the correct prediction:
```
Running example inference...
Text: Man With Fogged-Up Glasses Forced To Finish Soup Using Other Senses
Prediction: Onion
Confidence: 87.76%

Probabilities:
NotOnion: 12.24%
Onion: 87.76%
```