There are currently some performance hurdles that we suspect are caused by too much switching between cutensornet and python within hot paths. This might be alleviated somewhat if we avoid such behaviour by encapsulating the behaviour within C++ or cuda directly, and using pytket-cutensornet to wrap that functionality in a python interface.
There are currently some performance hurdles that we suspect are caused by too much switching between cutensornet and python within hot paths. This might be alleviated somewhat if we avoid such behaviour by encapsulating the behaviour within C++ or cuda directly, and using pytket-cutensornet to wrap that functionality in a python interface.