deployment

deployment

Deployment endpoint configuration.

DeploymentConfig dataclass

DeploymentConfig(checkpoint_path=MISSING, checkpoint_name=value, device=None, max_steps=1000000, compile_model=True, client=InferenceClientConfig())

Configuration for the real-time deployment endpoint.

Policy behavior comes from the checkpoint; these fields only describe the deployment environment and the action-execution strategy.

Attributes:

Name	Type	Description
checkpoint_path	str	Directory containing the checkpoint and its config.
checkpoint_name	str	Checkpoint filename inside checkpoint_path.
device	str | None	None selects cuda when available, else cpu.
max_steps	int	Maximum environment steps before the client stops.
compile_model	bool	Whether the policy is compiled with torch.compile.
client	InferenceClientConfig	Socket inference client settings.