Selective Rotary Position Embedding
This work addresses the challenge of position encoding in transformers, which is crucial for language modeling and sequence tasks, by proposing a novel input-dependent method that enhances performance, though it appears incremental as it builds upon existing RoPE and selective mechanisms.
The authors tackled the problem of encoding position information in transformers by introducing Selective RoPE, an input-dependent rotary embedding mechanism that generalizes RoPE and enables rotation in arbitrary angles for both linear and softmax transformers, demonstrating improved performance in language modeling and on difficult sequence tasks like copying, state tracking, and retrieval.
Position information is essential for language modeling. In softmax transformers, Rotary Position Embeddings (\textit{RoPE}) encode positions through \textit{fixed-angle} rotations, while in linear transformers, order is handled via input-dependent (selective) gating that decays past key-value associations. Selectivity has generally been shown to improve language-related tasks. Inspired by this, we introduce \textit{Selective RoPE}, an \textit{input-dependent} rotary embedding mechanism, that generalizes \textit{RoPE}, and enables rotation in \textit{arbitrary angles} for both linear and softmax transformers. We show that softmax attention already performs a hidden form of these rotations on query-key pairs, uncovering an implicit positional structure. We further show that in state-space models and gated linear transformers, the real part manages forgetting while the imaginary part encodes positions through rotations. We validate our method by equipping gated transformers with \textit{Selective RoPE}, demonstrating that its input-dependent rotations improve performance in language modeling and on difficult sequence tasks like copying, state tracking, and retrieval.