Discrete Flow Maps

arXiv stat.ML / 4/14/2026

📰 NewsSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper addresses a key bottleneck in autoregressive LLMs: their inherently sequential next-token generation limits speed, motivating alternative parallel generation methods.
It proposes Discrete Flow Maps, which compress the model’s generative trajectory into a single-step mapping to enable full-sequence generation from noise in one forward pass.
Unlike prior discrete flow approaches that used Euclidean regression losses ill-suited to discrete probability data, the method reformulates training to respect the geometry of the probability simplex.
By aligning the flow-map training dynamics with the discrete structure of language, the authors report improved empirical results, surpassing prior state-of-the-art in discrete flow modeling.

Abstract

The sequential nature of autoregressive next-token prediction imposes a fundamental speed limit on large language models. While continuous flow models offer a path to parallel generation, they traditionally demand expensive iterative integration. Flow Maps bypass this bottleneck by compressing generative trajectories into single-step mappings, theoretically enabling the generation of full text sequences from noise in a single forward pass. However, standard formulations rely on Euclidean regression losses that are geometrically ill-suited for discrete data. In this work, we resolve this conflict with Discrete Flow Maps, a framework that reconciles trajectory compression with the geometry of the probability simplex. We recast standard flow map training for the discrete domain, aligning the training dynamics with the discrete nature of language. Empirically, this strict geometric alignment allows our method to surpass previous state-of-the-art results in discrete flow modeling.