DeBERTa-v3-base

DeBERTa-v3-base is a pretrained English NLU encoder model from Microsoft Research with 86M backbone parameters, designed for fine-tuning on tasks like text classification and question answering. It improves on earlier BERT/RoBERTa-style models via disentangled attention and ELECTRA-style replaced-token-detection pretraining. Source: supabase.