Skip to main content
This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub
The SamplerDPMAdaptative node implements an adaptive DPM (Diffusion Probabilistic Model) sampler that automatically adjusts step sizes during the sampling process. It uses tolerance-based error control to determine optimal step sizes, balancing computational efficiency with sampling accuracy. This adaptive approach helps maintain quality while potentially reducing the number of steps needed.

Inputs

ParameterData TypeRequiredRangeDescription
orderINTYes2-3The order of the sampler method (default: 3)
rtolFLOATYes0.0-100.0Relative tolerance for error control (default: 0.05)
atolFLOATYes0.0-100.0Absolute tolerance for error control (default: 0.0078)
h_initFLOATYes0.0-100.0Initial step size (default: 0.05)
pcoeffFLOATYes0.0-100.0Proportional coefficient for step size control (default: 0.0)
icoeffFLOATYes0.0-100.0Integral coefficient for step size control (default: 1.0)
dcoeffFLOATYes0.0-100.0Derivative coefficient for step size control (default: 0.0)
accept_safetyFLOATYes0.0-100.0Safety factor for step acceptance (default: 0.81)
etaFLOATYes0.0-100.0Stochasticity parameter (default: 0.0)
s_noiseFLOATYes0.0-100.0Noise scaling factor (default: 1.0)

Outputs

Output NameData TypeDescription
samplerSAMPLERReturns a configured DPM adaptive sampler instance