Current backend implementation does not support multi-shot statistics. Instead, it requires the user to do each shot separately, incurring in the contraction cost for each shot.
When the user is only concern about the measurement statistics on a subset of the qubits, obtaining the reduced density matrix of the state on said qubits allows for efficient multi-shots (assuming few qubits are measured), and only incur in the cost of contraction once. CuTensorNet's high-level API provides a create_marginal method to obtain said reduced density matrix. We should make use of this to provide a multi-shot backend.
Current backend implementation does not support multi-shot statistics. Instead, it requires the user to do each shot separately, incurring in the contraction cost for each shot.
When the user is only concern about the measurement statistics on a subset of the qubits, obtaining the reduced density matrix of the state on said qubits allows for efficient multi-shots (assuming few qubits are measured), and only incur in the cost of contraction once. CuTensorNet's high-level API provides a
create_marginal
method to obtain said reduced density matrix. We should make use of this to provide a multi-shot backend.