aria1th
DeepCache seems to break TensorRT?
I have TensorRT working with SDXL (using dev branch of A1111). It works with SDXL_base in 768 and 1024. But if I activate your DeepCache extension, I get this error (known from when TensorRT is used with a wrong resolution, or not compiled with the dev branch of SD): "RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument mat1 in method wrapper_CUDA_addmm)" (at least, I am assuming this relates to TensorRT, but maybe it just happens to be the same error message?)
@arnolde Currently, the offloading option is not implemented, means DeepCache does not try to move tensors actively - its rather TensorRT side I suspect, the only move call related is here
If there was any error message related to specific part of the code, it will be more helpful.
