Generative module (AtomisticGenerator, hooks, pipelines)#

The generative API drives inference for generative models of any family — diffusion / flow matching, GANs, VAEs, normalizing flows — through the AtomisticGenerator driver: an optional condition step followed by generation, with lifecycle hooks, streaming, and sequential composition. For orientation and recipes, see the generative models user guide.

Core classes#

AtomisticGenerator

Generative inference driver: optionally condition the inputs, generate, optionally materialize to a Batch.

GeneratingFunction

Callable that encapsulates a family-specific generation strategy.

ConditionFunction

Translate a call's raw request into the generating function's working input.

GenerationStage

Stages of the AtomisticGenerator pipeline.

GenerationContext

Context object passed to generation hooks.

GenerationPipeline

Sequential composition of generation and batch-processing stages.

Model-side API#

GenerativeModelConfig

Pydantic config schema for a generative (non-energy) model.

GenerativeModelMixin

Mixin class for models designed for generative workflows.

Demo models#

DemoGANModel

Minimal GAN-side demo: a latent draw decoded to a point cloud.

DemoDiffusionModel

Minimal diffusion-side demo: an x0-predictor over point clouds.