DAPE: Data-Adaptive Positional Encoding for Length Extrapolation

TL;DR

This paper introduces DAPE, a data-adaptive positional encoding method for transformers that dynamically adjusts to input data, significantly improving length extrapolation and generalization capabilities on real-world datasets.

Contribution

We propose a novel DAPE method that adapts positional encoding based on input context, enhancing length generalization in transformer models beyond static encoding approaches.

Findings

DAPE improves performance on length extrapolation tasks.

DAPE outperforms static positional encoding methods at longer sequence lengths.

The model maintains local and anti-local information effectively.

Abstract

Positional encoding plays a crucial role in transformers, significantly impacting model performance and length generalization. Prior research has introduced absolute positional encoding (APE) and relative positional encoding (RPE) to distinguish token positions in given sequences. However, both APE and RPE remain fixed after model training regardless of input data, limiting their adaptability and flexibility. Hence, we expect that the desired positional encoding should be data-adaptive and can be dynamically adjusted with the given attention. In this paper, we propose a Data-Adaptive Positional Encoding (DAPE) method, which dynamically and semantically adjusts based on input context and learned fixed priors. Experimental validation on real-world datasets (Arxiv, Books3, and CHE) demonstrates that DAPE enhances model performances in terms of trained length and length generalization, where the improvements are statistically significant. The model visualization suggests that our model can keep both local and anti-local information. Finally, we successfully train the model on sequence length 128 and achieve better performance at evaluation sequence length 8192, compared with other static positional encoding methods, revealing the benefit of the adaptive positional encoding method.