Edit Flows: Flow Matching with Edit Operations

TL;DR

Edit Flows introduces a non-autoregressive sequence generation model using a discrete flow of edit operations, achieving superior performance in image captioning and text/code generation compared to existing models.

Contribution

The paper presents a novel non-autoregressive model that employs a continuous-time Markov chain over edit operations for flexible sequence generation.

Findings

Outperforms autoregressive and mask models in image captioning.

Significantly better than mask construction methods in text and code generation.

Efficient training via an expanded state space with auxiliary variables.

Abstract

Autoregressive generative models naturally generate variable-length sequences, while non-autoregressive models struggle, often imposing rigid, token-wise structures. We propose Edit Flows, a non-autoregressive model that overcomes these limitations by defining a discrete flow over sequences through edit operations$\unicode{x2013}$insertions, deletions, and substitutions. By modeling these operations within a Continuous-time Markov Chain over the sequence space, Edit Flows enable flexible, position-relative generation that aligns more closely with the structure of sequence data. Our training method leverages an expanded state space with auxiliary variables, making the learning process efficient and tractable. Empirical results show that Edit Flows outperforms both autoregressive and mask models on image captioning and significantly outperforms the mask construction in text and code generation.