Extended Data Table 2 Training parameters

From: Quantum circuit synthesis with diffusion models

  1. For the multi-qubit entanglement generation task (referred here as SRV), we use two training steps as explained in Methods, Training section. The fine-tuning and compilation trainings are done with circuits with fixed qubit numbers (either 9 and 10 for entanglement generation or 3 for unitary compilation). Hence, no special padding is needed in the space (qubit) dimension. Length padding is done per training batch. For all trainings we use a batch size of 256.
Back to article page