Inference Configs

Inference uses a small hierarchy of frozen dataclasses.

Public Imports

from artifex.configs import DiffusionInferenceConfig, InferenceConfig, ProteinDiffusionInferenceConfig

base_config = InferenceConfig(
    name="base_inference",
    checkpoint_path="{ARTIFEX_CHECKPOINT_ROOT}/latest",
    batch_size=4,
    num_samples=8,
    device="cpu",
)

Available Types

• InferenceConfig
• DiffusionInferenceConfig
• ProteinDiffusionInferenceConfig

get_inference_config() selects the narrowest supported type that matches the YAML shape. The retained protein_diffusion_inference asset therefore loads as ProteinDiffusionInferenceConfig.

Key Fields

Base inference:

• checkpoint_path
• output_dir
• batch_size
• num_samples
• device

Diffusion-specific additions:

• sampler
• timesteps
• temperature
• guidance_scale
• save_intermediate_steps

Protein-specific additions:

• target_seq_length
• backbone_atom_indices
• calculate_metrics
• visualize_structures
• save_as_pdb