Score: 0

Edit Flows: Flow Matching with Edit Operations

Published: June 10, 2025 | arXiv ID: 2506.09018v1

By: Marton Havasi , Brian Karrer , Itai Gat and more

Potential Business Impact:

Lets computers write better stories and code.

Business Areas:
Video Editing Content and Publishing, Media and Entertainment, Video

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-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.

Page Count
25 pages

Category
Computer Science:
Machine Learning (CS)