cirq.experiments.random_quantum_circuit_generation.get_random_combinations_for_device

For a given device, prepare a set of combinations to efficiently sample

parallel two-qubit XEB circuits.

n_library_circuits The number of circuits in your library. Likely the value passed to generate_library_of_2q_circuits.
n_combinations The number of combinations (with replacement) to generate using the library circuits. Since this function returns a CircuitLibraryCombination, the combinations will be represented by indexes between 0 and n_library_circuits-1 instead of the circuits themselves. The more combinations, the more precise of an estimate for XEB fidelity estimation, but a corresponding increase in the number of circuits you must sample.
device_graph A graph whose nodes are qubits and whose edges represent the possibility of doing a two-qubit gate. This combined with the pattern argument determines which two qubit pairs are activated when.
pattern A sequence of GridInteractionLayer, each of which has a particular set of qubits that are activated simultaneously. These pairs of qubits are deduced by combining this argument with device_graph.
random_state A random-state-like object to seed the random combination generation.

A list of CircuitLibraryCombination, each corresponding to an interaction layer in pattern where there is a non-zero number of pairs which would be activated. Each object has a combinations matrix of circuit indices of shape (n_combinations, len(pairs)) where len(pairs) may be different for each entry (i.e. for each layer in pattern). This returned list can be provided to sample_2q_xeb_circuits to efficiently sample parallel XEB circuits.