Towards More Efficient Insertion Transformer with Fractional Positional Encoding

TL;DR

This paper introduces Fractional Positional Encoding (FPE), a new method for Insertion Transformers that reuses representations to improve efficiency, reducing computation and latency in text generation tasks.

Contribution

The paper proposes FPE, a novel positional encoding scheme that enhances Insertion Transformers by enabling representation reuse, leading to more efficient parallel text generation.

Findings

FPE reduces floating-point operations in Insertion Transformers.

FPE improves latency during batched decoding.

Empirical results show FPE's effectiveness across various text generation tasks.

Abstract

Auto-regressive neural sequence models have been shown to be effective across text generation tasks. However, their left-to-right decoding order prevents generation from being parallelized. Insertion Transformer (Stern et al., 2019) is an attractive alternative that allows outputting multiple tokens in a single generation step. Nevertheless, due to the incompatibility between absolute positional encoding and insertion-based generation schemes, it needs to refresh the encoding of every token in the generated partial hypothesis at each step, which could be costly. We design a novel reusable positional encoding scheme for Insertion Transformers called Fractional Positional Encoding (FPE), which allows reusing representations calculated in previous steps. Empirical studies on various text generation tasks demonstrate the effectiveness of FPE, which leads to floating-point operation reduction and latency improvements on batched decoding.