jinaai / xlm-roberta-flash-implementation

This implementation is adapted from XLM-Roberta. In contrast to the original implementation, this model uses Rotary positional encodings and supports flash-attention 2.