skrample 0.4.0

Composable sampling functions for diffusion models

Skrample 0.4.0

Composable sampling functions for diffusion models

Status

Vertical slice, overtaking most diffusers features in quickdif

Feature Flags

• beta-schedule -> scipy : For the Beta() schedule modifier
• brownian-noise -> torchsde : For the Brownian() noise generator
• cdf-schedule -> scipy : For the SigmoidCDF() schedule
• diffusers-wrapper -> torch : For the diffusers integration module
• pytorch -> torch : For the pytorch module
  • pytorch.noise : Custom generators
• all : All of the above
• dev : For running tests/

Samplers

• Euler
• DPM
  • 1st order, 2nd order, 3rd order
  • SDE
• Adams/IPNDM
• UniP & UniPC
  • N order, limited to 9 for stability
  • Custom solver via other SkrampleSampler types
• SPC
  • Basic fully customizable midpoint corrector

Schedules

• Linear
• Scaled
  • uniform flag, AKA "trailing" in diffusers
• SigmaCDF
• ZSNR

Schedule modifiers

• Karras
• Exponential
• FlowShift
• Beta
• Hyper

Predictors

• Epsilon
• Velocity / vpred
• Flow

Noise generators

• Random
• Brownian
• Offset
• Pyramid

Integrations

Diffusers

• Compatibility for pipelines
  • SD1
  • SDXL
  • SD3
  • Flux
  • Others?
• Import from config
  • Sampler
  • Schedule
  • Predictor
• Manage state
  • Steps
  • Higher order
  • Generators
  • Config as presented

Implementations

quickdif

My diffusers cli quickdif has full support for all major Skrample features, allowing extremely fine-grained customization.