ConceptCTMC / Continuous-Time Markov Chain
Continuous-Time Markov Chain
A Markov process in which states jump at continuous time, serving as the mathematical framework for discrete diffusion models. Continuous-time Markov chains describe stochastic transitions between discrete states through rate matrices, enabling the formulation of forward and reverse diffusion processes over sequences of tokens.
Timeline (last 90 days)
Nov 6
Exact continuous-time sampling typically requires one model evaluation per token in sequence, so the number of evaluations is proportional to the number of tokens.
Unverified50%