Further workflow customisation

This user guide shows further workflow customisation methods, including flags, subflows, and task retries.

04_customising_workflows

This guide covers some of the same content from the 04_customising_workflows examples in JupyterLab. If you have access to Jupyter, it might be a better idea to follow the examples there, as this will give you hands on experience navigating the different files required to run a workflow.

Flags

Sometimes you might want to run a workflow without executing every task. This could be to speed up repeated runs, for debugging, or because a specific experiment isn't needed.

You can easily include or exclude specific tasks using flags (--flags). In the flow definition shown below, you can see that we're using two flags: run_on_flags and skip_on_flags.

name: onboarding-04-example-4-flags
qubits: [Q1, Q3]
couplings: []
stages:
  init:
    - name: schema
    - name: bootstrap-kb
      conditions:
        run_on_flags:
          - bootstrap
  experiments:
    qubit:
      - name: ramsey
      - name: time-t1
        dependencies:
          - ramsey
      - name: readout-discriminator-train
        dependencies:
          - time-t1
      - name: ramsey-ef
        dependencies:
          - readout-discriminator-train
        conditions:
          skip_on_flags:
            - no-ramsey-ef       
      - name: t2star-qpt-ramsey
        dependencies:
          - ramsey-ef
    coupling: []

As you might expect, the condition run_on_flags means the bootstrap-kb experiment will only run when the bootstrap flag is passed, i.e. when you run the command

qruise flow run --flags bootstrap

If you simply run, qruise flow run (or use any other flags), the bootstrap-kb experiment won't be executed.

Contrastingly, ramsey-ef has the condition skip_on_flags. To skip this experiment, you can use the command

qruise flow run --flags no-ramsey-ef

Without this flag, the ramsey-ef experiment will be executed.

Note

If an experiment task is dependent on a task that's skipped using a flag, it will inherit the dependency of the skipped task. For example, in this workflow, if ramsey-ef is skipped using the flag no-ramsey-ef, t2star-qpt-ramsey will inherit the dependency readout-discriminator-train.

Subflows

Subflows allow you to group experiments together. In fact, we've already been using subflows in the previous workflow user guides. If you look back at our earlier workflows, you'll see they're split into subflows such as init and experiment. This separates the initialisation tasks from the experiment tasks.

You can also define additional subflows to further organise your workflow. For example, grouping related experiments together, as shown in the code snippet below. This might be useful if you want to execute only a few experiments; for example, repeating a sequence until a certain condition is met, such as finding a suitable set of parameters or reaching a target fidelity.

The last experiment in the subflow can be used to check the condition and decide if the subflow should be repeated in case of a failure.

In the code snippet below, the subflow t-measurement is highlighted in blue. To define the subflow, you need to give it a name and then define the experiments as tasks the way you would with any other workflow.

name: onboarding-04-example-5-subflows
qubits: [Q1, Q3]
couplings: []
stages:
  init:
    - name: schema
    - name: bootstrap-kb
  experiments:
    qubit:
      - name: t-measurement
        tasks:
          - name: ramsey
          - name: time-t1
            dependencies:
              - ramsey
          - name: readout-discriminator-train
            dependencies:
              - time-t1
      - name: ramsey-ef
        dependencies:
          - t-measurement
      - name: t2star-qpt-ramsey
        dependencies:
          - t-measurement
    coupling: []

In the screen capture below, you can see that this workflow contains two subflows: init and experiment. Nested within the experiment subflow is the t-measurement subflow, which groups the ramsey, time-t1, and readout-discriminator-train experiment tasks.

Retry

Some tasks in your workflow might have a tendency to fail due to noise or other instabilities. In this case, you might want to retry the same task multiple times for a higher chance of success.

You can see in the screen capture above that the readout-discriminator-train task failed on both qubits. We can define a limited number of retries for this task, as shown in the code snippet below. If it succeeds within 4 retries, the workflow will proceed to the next task. If it still fails, the task will be marked as failed, and the subsequent tasks for that qubit will not be executed.

name: onboarding-04-example-6-retry
qubits: [Q1, Q3]
couplings: []
stages:
  init:
    - name: schema
    - name: bootstrap-kb
  experiments:
    qubit:
      - name: ramsey
      - name: time-t1
        dependencies:
          - ramsey
      - name: readout-discriminator-train
        dependencies:
          - time-t1
        retry:
          limit: 4    
      - name: ramsey-ef
        dependencies:
          - readout-discriminator-train
      - name: t2star-qpt-ramsey
        dependencies:
          - ramsey-ef
    coupling: []