Logger

torchrl.record.loggers.Logger(*args, use_ray_service=False, **kwargs)

A template for loggers.

Keyword Arguments:

• use_ray_service (bool) – If True, the logger runs as a Ray actor in a separate process. All method calls are delegated to the remote actor via ray.get(actor.method.remote(...)). CUDA tensors in log_metrics() and log_video() are automatically moved to CPU before the remote call. Requires ray to be installed. Default: False.

• ray_actor_options (dict, optional) – Options passed to ray.remote() when creating the Ray actor (e.g., {"num_cpus": 1}). Only used when use_ray_service=True.