celalkartoglu's picture
Upload README.md with huggingface_hub
f8f46c7 verified
# Türkçe Toksisite Sınıflandırma — BERTurk (5 sınıf)
**Etiketler:** DİĞER, KÜFÜR, HAKARET, IRKÇI, CİNSİYETÇİ
**Veri:** Overfit-GM/turkish-toxic-language (77,800 örnek)
**Taban model:** dbmdz/bert-base-turkish-cased
**Eğitim:** 3 epoch, bs=16/32, lr=3e-5, class_weight ile dengeleme
## Değerlendirme (Validation)
- Accuracy: 0.9563
- Macro-F1: 0.9185
- Loss: 0.4408
## Kullanım
```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TextClassificationPipeline
tok = AutoTokenizer.from_pretrained("celalkartoglu/tr-toxic-bert-multiclass-v1")
mdl = AutoModelForSequenceClassification.from_pretrained("celalkartoglu/tr-toxic-bert-multiclass-v1")
pipe = TextClassificationPipeline(model=mdl, tokenizer=tok, top_k=1)
pipe("Defol git buradan.") # -> [{'label': 'HAKARET', 'score': ...}]