nvalchemi.models.gen.GenerativeModelConfig#

class nvalchemi.models.gen.GenerativeModelConfig(*, required_inputs, outputs, prediction_outputs=None)[source]#

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

A GenerativeModelConfig is the contract between a generative model wrapper and the rest of nvalchemi. Every GenerativeModelMixin subclass must set a self.model_config instance in its __init__ (no class-level default, so each wrapper owns its own config object — mirroring BaseModelMixin).

The base schema holds only the three capability fields below. Models with model-specific runtime fields should subclass it (e.g. class ClariConfig(GenerativeModelConfig): temperature: float = 1.0) — the mixin’s isinstance enforcement accepts subclasses.

Parameters:
  • required_inputs (frozenset[str])

  • outputs (frozenset[str])

  • prediction_outputs (set[str] | None)

required_inputs#

Batch fields the model’s conditioning reads (empty means unconditional). Declared here so a GenerationPipeline can validate stage links at construction; a generator writes something by definition, so declarations are required, not optional.

Type:

Annotated[frozenset[str], Field(description=”Batch fields the model’s conditioning reads (empty = unconditional). Required: feeds GenerationPipeline link validation.”)]

outputs#

Batch fields the model’s generated output carries (written or forwarded). Distinct namespace from prediction_outputs, which keys ModelOutputs (e.g. "flow"), not batch fields.

Type:

Annotated[frozenset[str], Field(description=”Batch fields the model’s generated output carries (written or forwarded). Required: feeds GenerationPipeline link validation.”)]

prediction_outputs#

Output keys the model predicts (e.g. {"flow"}). None defaults to {"flow"} at use time.

Type:

Annotated[set[str] | None, Field(default=None, description=”Output keys the model predicts (e.g. {‘flow’}). None defaults to {‘flow’} at use time.”)]

Examples

>>> from nvalchemi.models.gen.base import GenerativeModelConfig
>>> cfg = GenerativeModelConfig(
...     required_inputs=frozenset({"positions", "atomic_numbers"}),
...     outputs=frozenset({"positions", "atomic_numbers", "cell"}),
... )
>>> cfg.prediction_outputs is None
True

Notes

extra="forbid": unknown constructor keywords raise pydantic.ValidationError, matching the ModelConfig pattern.

model_config = {'extra': 'forbid'}#

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].