nusabert-text-only-s2

Fine-tuned dari LazarusNLP/NusaBERT-base, hyperparameter dicari dengan Optuna.

• Problem type: multi_label_classification
• Labels: insults, identity_attack, threat_incitement_to_violence, profanity_obscenity, sexually_explicit

Penggunaan

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("watermelonlalala/nusabert-text-only-s2")
model = AutoModelForSequenceClassification.from_pretrained("watermelonlalala/nusabert-text-only-s2")

Catatan: teks input sebaiknya diproses dengan preprocessing yang sama seperti saat training (lowercase, normalisasi emoji, mention, URL, dst.) agar hasil prediksi konsisten.