{\sigma}-GPTs: A New Approach to Autoregressive Models

TL;DR

This paper introduces -GPTs, a flexible autoregressive model that allows dynamic token sampling and conditioning, reducing generation steps significantly across multiple domains.

Contribution

It proposes a novel method to modulate token generation order on-the-fly using positional encoding, enabling more efficient and versatile sequence modeling.

Findings

Reduces number of steps for sequence generation by an order of magnitude.

Enables sampling of arbitrary token subsets and dynamic multi-token generation.

Demonstrates effectiveness across language, path-solving, and aircraft prediction tasks.

Abstract

Autoregressive models, such as the GPT family, use a fixed order, usually left-to-right, to generate sequences. However, this is not a necessity. In this paper, we challenge this assumption and show that by simply adding a positional encoding for the output, this order can be modulated on-the-fly per-sample which offers key advantageous properties. It allows for the sampling of and conditioning on arbitrary subsets of tokens, and it also allows sampling in one shot multiple tokens dynamically according to a rejection strategy, leading to a sub-linear number of model evaluations. We evaluate our method across various domains, including language modeling, path-solving, and aircraft vertical rate prediction, decreasing the number of steps required for generation by an order of magnitude.