| Published | 20 April 2021 |
|---|---|
| Authors | Jianlin Su, Yu Lu, Shengfeng Pan, Ahmed Murtadha, Bo Wen, Yunfeng Liu |
| arXiv Abstract | [[1]] |
| arXiv PDF | [[2]] |
The paper RoFormer: Enhanced Transformer with Rotary Position Embedding introduced the concept of rotary position embedding for transformers. This was introduced as an alternative to the concept of sinusoidal encoding, which was proposed in the paper Attention is All You Need that introduced transformers.
Abstract[edit | edit source]
Position encoding recently has shown effective in the transformer architecture. It enables valuable supervision for dependency modeling between elements at different positions of the sequence. In this paper, we first investigate various methods to integrate positional information into the learning process of transformer-based language models. Then, we propose a novel method named Rotary Position Embedding (RoPE) to effectively leverage the positional information. Specifically, the proposed RoPE encodes the absolute position with a rotation matrix and meanwhile incorporates the explicit relative position dependency in self-attention formulation. Notably, RoPE enables valuable properties, including the flexibility of sequence length, decaying inter-token dependency with increasing relative distances, and the capability of equipping the linear self-attention with relative position encoding. Finally, we evaluate the enhanced transformer with rotary position embedding, also called RoFormer, on various long text classification benchmark datasets. Our experiments show that it consistently overcomes its alternatives. Furthermore, we provide a theoretical analysis to explain some experimental results.