Training Configuration

pydantic model dpeva.config.TrainingConfig

Configuration for Training Workflow.

field allow_experimental: bool = False

Explicitly allow experimental data-format capabilities such as fine-tuning from deepmd/lmdb.

field base_model_path: Path [Required]

Path to base model.

field input_json_path: Path = PosixPath('input.json')

field model_head: str [Required]

Model head name.

field num_models: int = 4

Number of models (>=3 for UQ).

field seeds: List[int] | None = None

field template_path: Path | None = None

field training_data_path: Path | None = None

field training_mode: Literal['init', 'cont'] = 'init'

field training_seeds: List[int] | None = None