monologg/KoELECTRA
A Korean ELECTRA language model trained on 34GB of Korean text, available in base and small variants.

Velocity · 7d
+0.3
★ / day
Trend
→steady
star history
KoELECTRA is a pretrained ELECTRA-based language model specifically designed for Korean language processing. It uses replaced token detection training, which provides learning signals from all input tokens rather than just masked positions. The model comes in base and small variants, supports both PyTorch and TensorFlow frameworks, and integrates with the Hugging Face Transformers library for easy loading and inference.