Flexible Language Modeling in Continuous Space with Transformer-based Autoregressive Flows
This work addresses the need for more flexible language modeling for AI researchers, though it is incremental as it builds on existing autoregressive and normalizing flow methods.
The authors tackled the problem of language modeling's reliance on discrete tokens and unidirectional context by shifting to a continuous latent space, resulting in a framework that achieves strong likelihood performance on benchmarks and enables flexible features like bi-directional context and multi-pass generation.
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.