Asynchronous Flow
A continuous-time generative model framework that assigns an independent timeline to each token or dimension, allowing different positions to complete denoising at different rates, thereby encoding the generation order.
Timeline (last 90 days)
Asynchronous flow draws on the idea of flow matching from continuous-time generative models, in which each token has an independent time axis, allowing different positions to complete denoising at different rates.
Asynchronicity is implemented mathematically by defining a monotonically increasing time reparameterization function for each dimension.
The paper formalizes a general class of asynchronous flows that supports both prespecified schedules and schedules learned from data and adaptively adjusted with state.
The authors prove that such asynchronous flows can define valid flow maps and present a new flow-map distillation method.
All Facts (4)
Asynchronous flow draws on the idea of flow matching from continuous-time generative models, in which each token has an independent time axis, allowing different positions to complete denoising at different rates.
50%UnverifiedAsynchronicity is implemented mathematically by defining a monotonically increasing time reparameterization function for each dimension.
50%UnverifiedThe paper formalizes a general class of asynchronous flows that supports both prespecified schedules and schedules learned from data and adaptively adjusted with state.
50%UnverifiedThe authors prove that such asynchronous flows can define valid flow maps and present a new flow-map distillation method.
50%