Edit Flows: Variable Length Discrete Flow Matching with Sequence-Level Edit Operations
autoregressive modelsauxiliary variablescontinuous-time markov chaindeletionsdiscrete flowedit operationsempirical resultsimage captioninginsertionslearning processnon-autoregressive modelsposition-relative generationsequence spacestate spacesubstitutions
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