ConceptConsistency Training / 一致性蒸馏 / Consistency Distillation
Consistency Training
A training method for few-step sampling in diffusion models. It defines a consistency function that requires the model to map any point along the same denoising trajectory to the same initial state, enabling single-step or few-step sampling. Representative methods include Consistency Models.
Timeline (last 90 days)
Oct 8
Distillation, consistency training, and adversarial objectives are the mainstream approaches to few-step generation, but they sacrifice the likelihood framework in the process of compressing the number of steps
Unverified50%