Add methods to serialize multi-programs in cirq_google #7678
Conversation
dstrain115
commented
Sep 30, 2025
dstrain115
commented
- Adds a new Message in the Program proto for KeyedCircuit which can store multiple circuits indexed by parameters.
- This also adds a pair of functions in CircuitSerializer called serialize_multi_program and deserialize_multi_program.
- These can serialize a list of circuits, a map of circuits with string keys, or a function that returns a circuit (with a corresponding sweep object that specifies how to call the function).
- Adds a new Message in the Program proto for KeyedCircuit which can store multiple circuits indexed by parameters. - This also adds a pair of functions in CircuitSerializer called serialize_multi_program and deserialize_multi_program. - These can serialize a list of circuits, a map of circuits with string keys, or a function that returns a circuit (with a corresponding sweep object that specifies how to call the function).
dstrain115
requested review from
a team,
verult,
vtomole and
wcourtney
as code owners
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #7678 +/- ##
========================================
Coverage 99.37% 99.38%
========================================
Files 1085 1089 +4
Lines 96937 97526 +589
========================================
+ Hits 96331 96925 +594
+ Misses 606 601 -5 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
There was a problem hiding this comment.
I wonder if the creation of circuits from sweep parameters and a circuit factory should be moved to a separate function so we could simplify what argument types are accepted in serialize_multi_program. The type signature there could also accept Iterable[AbstractCircuit] and Iterable[tuple[str, AbstractCircuit]] so that serialize_multi_program could take a generator function output.
| results. | ||
| Raises: | ||
| NotImplementedError: If the program is of a type that is supported. |
There was a problem hiding this comment.
typo: supported --> not supported
There was a problem hiding this comment.
Eh, somehow it is still there on lines 136 and 139.
| | Callable[..., cirq.AbstractCircuit] | ||
| | Callable[..., Mapping[str, cirq.AbstractCircuit]] |
There was a problem hiding this comment.
IIUC, the callable arguments are used to generate circuits from the sweep parameters, essentially translating them to either a sequence or mapping of circuits, ie, the other two parameter types. The sweep argument is not serialized at all, it is only used to provide arguments for the callable.
How about factoring the circuit generation to a separate function, say circuits_from_sweep which would return either a sequence or mapping of Circuits?
If memory overhead is of concern, circuits_from_sweep could be a generator function and the multi_program argument here could be adjusted to accept Iterable[AbstractCircuit] | Iterable[tuple[str, AbstractCircuit]].
There was a problem hiding this comment.
I separated this out to serialize_circuit_function(). I don't think having circuits_from_sweep is worth it, since that is not intended functionality. You'd just want to serialize them immediately. What do you think?
There was a problem hiding this comment.
Looks good, thanks! I just realized that serialize_circuit_function also serializes param_tuple which is not available in serialize_multi_program (ie, they cannot be chained).
| raise ValueError(f'Function returned unrecognized type: {type(circuit_or_map)}') | ||
| for key, circuit in circuit_tuples: | ||
| new_program = msg.keyed_circuits.add() | ||
| if key is not None: |
There was a problem hiding this comment.
Nit - this should be always True looking at the code above. Maybe remove?
| ("", circuit_or_map) | ||
| ] | ||
| elif isinstance(circuit_or_map, Mapping): | ||
| circuit_tuples = list(circuit_or_map.items()) |
There was a problem hiding this comment.
circuit_tuples from different sweep points can have duplicate keys, so the Program.keyed_circuits sequence may get non-unique KeyedCircuit.key values as well.
If that is not an issue, perhaps the serialize_multi_program should also accept an iterable of (key, Circuit) pairs?
There was a problem hiding this comment.
I'm not totally sure I understand the concern. The combination of key and args should always be unique. I am not sure that the key needs to be unique on its own.
There was a problem hiding this comment.
I see - there is no requirement for unique keys. NVM, the initial comment, it is OK as is.
cirq-google/cirq_google/serialization/circuit_serializer_test.py
Outdated
Show resolved
Hide resolved
| constants=msg.constants, | ||
| raw_constants=raw_constants, | ||
| ) | ||
| elif isinstance(multi_program, Sequence): |
There was a problem hiding this comment.
Nit - Sequence --> Iterable or allow only Sequence[cirq.Circuit] in multi_program annotation. As it is, generator would pass type check, but raise an exception at runtime.
2 participants
