SentimentArEng
This model is a fine-tuned version of cardiffnlp/twitter-xlm-roberta-base-sentiment on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.502831
- Accuracy: 0.798512
inference with pipeline
from transformers import pipeline
model_path = "Noor0/SentimentArEng"
sentiment_task = pipeline("sentiment-analysis", model=model_path, tokenizer=model_path)
sentiment_task("تعامل الموظفين كان أقل من المتوقع")
- output:
- [{'label': 'negative', 'score': 0.9905518293380737}]
Training and evaluation data
- Training set: 114,885 records
- evaluation data: 12,765 records
Training procedure
| Training Loss | Epoch | Validation Loss | Accuracy |
|---|---|---|---|
| 0.4511 | 2.0 | 0.502831 | 0.7985 |
| 0.3655 | 3.0 | 0.576118 | 0.7954 |
| 0.3019 | 4.0 | 0.625391 | 0.7985 |
| 0.2466 | 5.0 | 0.835689 | 0.7979 |
Training hyperparameters
- The following hyperparameters were used during training:
- learning_rate=2e-5
- num_train_epochs=20
- weight_decay=0.01
- batch_size=16,
Framework versions
- Transformers 4.35.0
- Pytorch 2.0.0
- Datasets 2.11.0
- Tokenizers 0.14.1