nvalchemi-model-wrapping

What problem does it solve?

This approach lets you wrap any MLIP with the BaseModelMixin to standardize how models receive inputs, generate outputs, and produce embeddings within nvalchemi.

Core Features & Use Cases

• Standardized input/output interface via BaseModelMixin for arbitrary MLIPs.
• Clear embedding extraction and graph-level representations, plus export options for integration with external tools.
• Supports single AtomicData or Batch workflows and configurable runtime behavior through ModelCard/ModelConfig.
• Use case: wrap a PyTorch MLIP to produce energies and forces in a workflow, and easily export a compiled model for ASE calculators.

Quick Start

Create a wrapper by combining your PyTorch model with BaseModelMixin, implement adapt_input, adapt_output, and forward as demonstrated, then set model.model_config to enable desired computations.