Modern quantum circuit simulation is not just “machine learning with complex tensors.” It involves irregular tensor contractions, sparse operators, statevector transformations, and automatic differentiation through all of them. This makes backend choice unusually important. A backend that is excellent for standard neural-network layers may still be a poor fit for general quantum simulation workloads.
We benchmarked this with a simple VQE workload for the 1D transverse-field Ising
model as in the script,
H = -sum_i Z_i Z_{i+1} - sum_i X_i,
Enter fullscreen mode
