[Frontend] A prototype for one_shot_dynamic transform #734
Conversation
where `m` is a measurement result.
| qreg_in = qreg_out | ||
| # (1.5) If user wanted the `dynamic_one_shot` transformation we apply it by re-tracing the | ||
| # function using the wrapper loop. | ||
| if qnode.needs_dynamic_one_shot and device.shots.total_shots != 1: |
There was a problem hiding this comment.
dynamic_one_shot may also be applied to a single shot tape, but we can neglect that case at first I would say.
|
Hello. You may have forgotten to update the changelog!
|
| @for_loop(0, orig_shots, 1) | ||
| def loop(i, s): | ||
| r = f(*args2, **kwargs2) | ||
| s = s.at[i].set(r) |
There was a problem hiding this comment.
It appears r comes out as a quantum measurements (e.g. qml.expval) instead of the results (e.g. 0.0) which throws off the rest of the tracing. Is it just me?
There was a problem hiding this comment.
Hmm, I did not pay attention to this. My idea is that it might be a correct resulting tracer of some kind. In the example I used catalyst.measure (returning a boolean tracer) but I hoped that measurements would just return a measurement tracer. The thing is - we do not know its value at the compile time.
[sc-63049]
See also an internal document regarding this issue.