Back to Openvino

Write Inference Config

docs/articles_en/physical-ai/how-to/config/write-inference-config.md

2026.2.1764 B
Original Source

Write Inference Config

Preview: The config system (physicalai.config) is a planned API. The examples below document the target design.

Use an inference config when you need to author an inference pipeline outside an exported manifest.

yaml
model:
  class_path: physicalai.inference.InferenceModel
  init_args:
    export_dir: ./exports/act_policy
    backend: openvino
    device: CPU

If the manifest already contains the required runner, artifacts, processors, and hardware metadata, prefer loading from the manifest instead.

python
model = InferenceModel.load("./exports/act_policy")
action = model.select_action(observation)

Use workflow config to express user-authored intent. Use a manifest to describe exported package metadata.