MemoryError raised by
DistributedRunner.
Development preview. See availability.
/dev/shm.
Both a single worker and multiple workers use this transport.
For local containers, follow the shared-memory instructions in
Test locally. For deployed models, check
the deployment’s shared-memory capacity and memory limit. A local --shm-size setting does not
change a deployed container.
An input allocation can fail before any worker receives the request. A result allocation can fail
inside a worker. Do not assume the call left model state unchanged. Check runner health and the
model’s recovery policy before another call.