DAPE: Data-Adaptive Positional Encoding for Length Extrapolation
The Chinese University of Hong Kong · National University of Singapore · Hong Kong University of Science and Technology · Chinese University of Hong Kong · University of Hong Kong · Noah's Ark Lab · Hong Kong Baptist University · Noah’s Ark Lab, Huawei Technologies · Noah's Ark Lab, Huawei Tech Investment Co Ltd · Department of Computer Science and Engineering, The Chinese University of Hong Kong
PDF 由论文原始站点提供，PaperCompass 不保存论文文件。
摘要
Positional encoding plays a crucial role in transformers, significantly impact- ing 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.