(but not quite)

About this backend:

• General-use quantum information library (QUIMB) repurposed for a GPU-accelerated tensor network backend for medium-large scale noisy quantum circuit simulation.
• Supports NVIDIA GPUs, tested on Lambda workstation (2x NVIDIA RTX A6000).
• Requires cupy to interface between Python and CUDA.
• Written in Python, slow on CPU-only nodes (compared to C/C++ backends).
• Uses opt_einsum format for contraction path logic; can be interfaced with cuQuantum/cuTensorNet.

NOTES:

• Offloading matrix multiplications to GPU has very little overhead with the right preparation, and allows them to be executed much more quickly.
• There is currently an issue with the copy() function in the TN classes, especially when doing a virtual copy. This copy function is pointed to deepcopy in all inherited classes to circumvent this issue; however, this slows down sampling and it may be necessary to implement virtual TN handling within the child classes as well.

The official documentation of QUIMB, the original software package, can be found at: http://quimb.readthedocs.io/en/latest/.