DAPE: Data-Adaptive Positional Encoding for Length Extrapolation
This addresses the issue of length extrapolation in transformers for NLP tasks, offering an incremental improvement over existing positional encoding methods.
The paper tackled the problem of fixed positional encodings in transformers limiting adaptability and length generalization by proposing DAPE, a data-adaptive method that dynamically adjusts based on input context, achieving better performance at evaluation sequence length 8192 after training on length 128 compared to static methods.
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.