I've tried to compile the model with the new torch_tensorrt torchscript compiler.
However, it does not seem to support a variable number of arguments which is heavily used for the various forward() functions in the model definition:
torch.jit.frontend.NotSupportedError: Compiled functions can't take variable number of arguments or use keyword-only arguments with defaults:
File "/home/frederik/.local/lib/python3.8/site-packages/performer_pytorch/performer_pytorch.py", line 576
def forward(self, x, **kwargs):
~~~~~~~ <--- HERE
if self.auto_check_redraw:
self.proj_updater.redraw_projections()
If I were to make a pull request which removes the **kwargs in order to provide TensorRT and torchscript support, would it be merged?
Hi
Thank you for contributing this amazing repo!
I've tried to compile the model with the new torch_tensorrt torchscript compiler. However, it does not seem to support a variable number of arguments which is heavily used for the various forward() functions in the model definition:
If I were to make a pull request which removes the **kwargs in order to provide TensorRT and torchscript support, would it be merged?