The current version of the system does not execute properly when a GPU is forced. As structured the ArgLab WO server loads the workbench pipeline as Spacy then allows it to select or prefer the method. In general this does not select the GPU ion part because we have incompatible data structures in place. In order to fix this what we need to do is to investigate whether we can force the cuda data structures even without the GPU (and without substantial cost) and investigate whether we can implement a replacement model that will run both with and without GPUs.
The current version of the system does not execute properly when a GPU is forced. As structured the ArgLab WO server loads the workbench pipeline as Spacy then allows it to select or prefer the method. In general this does not select the GPU ion part because we have incompatible data structures in place. In order to fix this what we need to do is to investigate whether we can force the cuda data structures even without the GPU (and without substantial cost) and investigate whether we can implement a replacement model that will run both with and without GPUs.