Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated Distillation

TL;DR

This paper introduces TS-DFM, a guided navigation approach that improves discrete flow matching by evaluating and selecting better continuations during training, resulting in faster and more accurate language generation.

Contribution

TS-DFM replaces blind stochastic jumps with guided energy-based navigation, significantly enhancing discrete flow matching efficiency and performance without increasing inference cost.

Findings

Shaped student model achieves 32% lower perplexity than the teacher.

TS-DFM outperforms all compared discrete-generation baselines.

Model trained on 170M parameters achieves state-of-the-art perplexity.

Abstract

Discrete flow matching generates text by iteratively transforming noise tokens into coherent language, but may require hundreds of forward passes. Distillation uses the multi-step trajectory to train a student to reproduce the process in a few steps. When the student underperforms, the usual explanation is insufficient capacity. We argue the opposite: the trajectory is the bottleneck, not the student. Each training trajectory is built through a chain of blind stochastic jumps with no evaluation of sequence quality; a single bad decision at an early midpoint propagates through subsequent steps, yet the student must imitate the result. Trajectory-Shaped Discrete Flow Matching (TS-DFM) replaces these blind jumps with guided navigation: a lightweight energy compass evaluates candidate continuations at each midpoint, selecting the most coherent. All shaping is training-only; inference cost is unchanged. On 170M-parameter language modeling, the shaped student at 8 steps achieves 32% lower perplexity than the 1,024-step teacher while being 128x faster, with gains consistent across source distributions and three evaluators of increasing scale. TS-DFM achieves the best perplexity of any discrete-generation baseline we compare against, including methods trained on 6x more data or using 5x larger models.