Botocore ResourceNotFoundException while running on AWS Braket SV1

#1

I am trying to run a 16-qubit circuit on the AWS Braket SV1 simulator through Pennylane. My qml device uses parallel=True. I am getting the following message although I can find the quantum task XX folder on amazon S3:

INFO:backoff:Backing off get_quantum_task(...) for 1.0s (botocore.errorfactory.ResourceNotFoundException: An error occurred (ResourceNotFoundException) when calling the GetQuantumTask operation: Quantum task XX not found)

Besides, the simulation is slower on SV1 than on local simulator (i.e., on qml.device('default.qubit', wires=n_qubits)).

(1) Is the slow simulation on SV1 related to the exception?
(2) Is there a remedy from pennylane’s end or do I need to look for solutions in Braket end?

Thanks.

#2

Hey @asaki!

I am getting the following message although I can find the quantum task XX folder on amazon S3

This error message comes from the Braket/AWS stack, so you may want to discuss directly with their team (e.g., by posting on their GitHub issues page). However, I can say from my experience that this warning can occur occasionally and didn’t cause me any problems.

Besides, the simulation is slower on SV1 than on local simulator

Yes, the remote SV1 device is not always faster than local devices. SV1 really starts to shine when:

  • The number of qubits goes above 20. The reason that default.qubit can be faster for smaller qubit numbers is because it’s being run locally, whereas we have to pay the cost of sending jobs to a remote server when running on SV1. Check out our tutorial to understand more.

  • Gradient calculations can be performed in parallel on SV1 (when the device is loaded with parallel=True), so you are likely to see the greatest speedups on SV1 when performing backwards passes/differentiation rather than forwards passes/evaluations.

Thanks!