SiT
An important variant of DiT that generalizes discrete noise schedules into a continuous-time interpolation framework, making sampling and training more flexible and easier to scale.
Timeline (last 90 days)
The paper "Representation Conditioning for Generation" proposes the RECON method, which, without adding any parameters, speeds up diffusion Transformer training convergence to 63x that of SiT and 25x that of REPA.
SiT (Scalable Interpolant Transformers) is an important variant of DiT that generalizes discrete noise schedules into a continuous-time interpolation framework, making sampling and training more flexible and easier to scale.
All Facts (2)
The paper "Representation Conditioning for Generation" proposes the RECON method, which, without adding any parameters, speeds up diffusion Transformer training convergence to 63x that of SiT and 25x that of REPA.
50%UnverifiedSiT (Scalable Interpolant Transformers) is an important variant of DiT that generalizes discrete noise schedules into a continuous-time interpolation framework, making sampling and training more flexible and easier to scale.
50%