MonarchTransport

class torchrl.modules.inference_server.MonarchTransport(*, max_queue_size: int = 1000)

Transport using Monarch for distributed inference on GPU clusters.

Uses Monarch’s actor model and RDMA-capable channels for efficient cross-node communication. Monarch is imported lazily at instantiation time; importing the class itself does not require Monarch.

Note

This transport requires monarch to be installed. It is designed for large-scale GPU clusters where Monarch is the preferred communication layer.

Keyword Arguments:

max_queue_size (int) – maximum size of the request queue. Default: 1000.

client() → _QueueInferenceClient

Create an actor-side client with a dedicated response queue.

Returns:

A _QueueInferenceClient that can be passed to a Monarch actor.