SeqPE: Transformer with Sequential Position Encoding
This addresses the adaptability and scalability challenges of positional encodings for Transformers, enabling better extrapolation and multi-dimensional generalization, though it is incremental as it builds on existing encoding methods.
The authors tackled the problem of positional encodings in Transformers being limited in extrapolation and adaptability to new modalities, and introduced SeqPE, a fully learnable framework that represents positions as sequences and uses a sequential encoder with regularization objectives. Experiments showed SeqPE surpasses baselines in perplexity, exact match, and accuracy under context length extrapolation, with improvements like 2.1% higher EM on long-context QA and seamless generalization to multi-dimensional inputs.
Since self-attention layers in Transformers are permutation invariant by design, positional encodings must be explicitly incorporated to enable spatial understanding. However, fixed-size lookup tables used in traditional learnable position embeddings (PEs) limit extrapolation capabilities beyond pre-trained sequence lengths. Expert-designed methods such as ALiBi and RoPE, mitigate this limitation but demand extensive modifications for adapting to new modalities, underscoring fundamental challenges in adaptability and scalability. In this work, we present SeqPE, a unified and fully learnable position encoding framework that represents each $n$-dimensional position index as a symbolic sequence and employs a lightweight sequential position encoder to learn their embeddings in an end-to-end manner. To regularize SeqPE's embedding space, we introduce two complementary objectives: a contrastive objective that aligns embedding distances with a predefined position-distance function, and a knowledge distillation loss that anchors out-of-distribution position embeddings to in-distribution teacher representations, further enhancing extrapolation performance. Experiments across language modeling, long-context question answering, and 2D image classification demonstrate that SeqPE not only surpasses strong baselines in perplexity, exact match (EM), and accuracy--particularly under context length extrapolation--but also enables seamless generalization to multi-dimensional inputs without requiring manual architectural redesign. We release our code, data, and checkpoints at https://github.com/ghrua/seqpe.