Runtime increases when training NN with qiskit qasm_simulator

Hi. I’m trying to build a Hybrid NN with pennylane. It works fine with default.qubits device. However, when trying to access IBM’s qasm_simulator, each epoch takes a lot of time. I get this warning inside every epoch multiple times which is not observed in qiskit.aer simulator or default.qubits.

WARNING:websocket:websocket connected

I want to use IBM’s hardware. Is there something I’m missing that has to be taken care??

Hi @Ganesh_Chockalingam, thank you for posting this here. I’m able to reproduce your problem. It seems like the problem is generated on IBM’s side but not for every user. We’re looking into this to try to figure out the problem. Could you please post the output of qml.about()?

Hello @CatalinaAlbornoz. Sorry for the late reply. Here is the output of qml.about()

I’ve mentioned execution time per epoch for more clarity.

  1. default.qubits - 14 sec
  2. qiskit.aer simulator - 126 sec
  3. qiskit.ibmq(‘ibmq_qasm_simulator’) - more than an 1 hr (And I get that warning multiple times)

Hi @Ganesh_Chockalingam ,

Thank you very much for sharing these additional details. I’ve checked with our latest version of PennyLane and the warning is still appearing.

The warning really just means that there has been a connection so it’s not necessarily a bad thing. The speed however is quite slow.

I’m thinking that you may get better speeds if you try using runtime.

The conversation here might be helpful to get started with it.

Please let me know if you have any additional questions!