Flexible Language Modeling in Continuous Space with Transformer-based Autoregressive Flows

TL;DR

This paper introduces TarFlowLM, a novel transformer-based autoregressive normalizing flow framework that models language in a continuous latent space, offering enhanced flexibility such as bi-directional context, block-wise generation, and hierarchical multi-pass decoding.

Contribution

The work presents a new continuous-space language modeling paradigm with transformer-based autoregressive flows, enabling flexible context modeling and generation strategies beyond traditional discrete token approaches.

Findings

Achieves strong likelihood performance on benchmarks.

Demonstrates flexible global and local context modeling.

Supports hierarchical and block-wise generation.

Abstract

Autoregressive models have driven remarkable progress in language modeling. Their foundational reliance on discrete tokens, unidirectional context, and single-pass decoding, while central to their success, also inspires the exploration of a design space that could offer new axes of modeling flexibility. In this work, we explore an alternative paradigm, shifting language modeling from a discrete token space to a continuous latent space. We propose a novel framework TarFlowLM, that employs transformer-based autoregressive normalizing flows to model these continuous representations. This approach unlocks substantial flexibility, enabling the construction of models that can capture global bi-directional context through stacked, alternating-direction autoregressive transformations, support block-wise generation with flexible token patch sizes, and facilitate a hierarchical multi-pass generation process. We further propose new mixture-based coupling transformations designed to capture complex dependencies within the latent space shaped by discrete data, and demonstrate theoretical connections to conventional discrete autoregressive models. Extensive experiments on language modeling benchmarks demonstrate strong likelihood performance and highlight the flexible modeling capabilities inherent in our framework.