Coupling Models for One-Step Discrete Generation

2026-05-07

AI Summary

Efficient one-shot generation of discrete structures has long been hindered by autoregressive or iterative approaches. This work proposes an end-to-end coupled model that directly links discrete sequences to Gaussian latent variables via a learnable mapping and introduces a dedicated single-step decoder, eschewing conventional paradigms reliant on complex continuous flows or predefined noise couplings. The method establishes, for the first time, a coupling mechanism explicitly tailored for single-step discrete generation. Empirical results demonstrate significant improvements: a 33% reduction in perplexity on LM1B text generation, an 18% gain in Fly Brain enhancer design (FBD) performance, and a 46% improvement in FID for binarized MNIST image generation.

Abstract

Generative modeling over discrete structures underpins applications across deep learning, from biological sequence design and code generation to large language models, yet generation often remains sequential, relying on autoregressive decoding or iterative refinement. In this work, we introduce Coupling Models(Coupling Models), a one-step discrete generative model that learns a direct coupling between discrete sequences and Gaussian latents. Unlike recent distillation methods that compress a pretrained multi-step sampler into a few steps, Coupling Model trains a purpose-built decoder to invert this coupling and generate samples in a single step. The model also avoids complex continuous flows over the simplex and hand-specified data-to-noise couplings. Empirically,Coupling Model improves the strongest one-step baselines in each domain: it reduces LM1B text-generation perplexity by 33% at its lowest-perplexity operating point, Fly Brain enhancer-design FBD by 18%, and MNIST-Binary FID by 46%. These results suggest that effective one-step discrete generation depends strongly on how data and noise are coupled before decoding. Code is available at https://github.com/pengzhangzhi/Coupling-Models.

Problem

• discrete generation
• one-step generation
• generative modeling
• autoregressive decoding
• iterative refinement

Innovation

• one-step generation
• discrete generative modeling
• Gaussian latent coupling
• non-autoregressive decoding
• invertible coupling