inference_client

inference_client

Shared configuration for the socket inference client.

InferenceClientConfig dataclass

InferenceClientConfig(model_server_address='127.0.0.1', model_server_port=5555, temporal_aggregation=False, action_execution_horizon=None, update_rate_hz=None, temporal_max_timesteps=800, timing_log=False, compression_type=value, request_timeout_seconds=None)

Settings for a client driving an environment server.

Used by the deployment endpoint and by online-inference explanations, which run the same client loop.

Attributes:

Name	Type	Description
model_server_address	str	Environment server address.
model_server_port	int	Environment server port.
temporal_aggregation	bool	Whether predicted chunks are ensembled.
action_execution_horizon	int | None	Actions executed per chunk; None executes the full prediction horizon.
update_rate_hz	float | None	Action send rate; None for simulation.
temporal_max_timesteps	int	Timesteps tracked by temporal aggregation.
timing_log	bool	Whether the client logs per-step timing.
compression_type	str	Wire compression for camera observations.
request_timeout_seconds	float | None	Per-request transport timeout; None blocks indefinitely.