ray.rllib.evaluation.rollout_worker.RolloutWorker.get_weights#

RolloutWorker.get_weights(policies: Collection[str] | None = None, inference_only: bool = False) → Dict[str, dict][source]#

Returns each policies’ model weights of this worker.

Parameters:

  • policies – List of PolicyIDs to get the weights from. Use None for all policies.

  • inference_only – This argument is only added for interface consistency with the new api stack.

Returns:

Dict mapping PolicyIDs to ModelWeights.

from ray.rllib.evaluation.rollout_worker import RolloutWorker
# Create a RolloutWorker.
worker = ...
weights = worker.get_weights()
print(weights)

{"default_policy": {"layer1": array(...), "layer2": ...}}