just stumble upon your repo when trying to use real-esrgan with tensorrt, i think the instructions can be a bit better, especially windows and the .bat file
install torch
if cuda 11.8: pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
if cuda 12.1: pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121
install tensorrt
if linux: pip install tensorrt
if windows: pip install + the tensorrt .whl file in the folder downloaded when install TensorRT
no need to install onnx_graphsurgeon nor tensorrt_dispatch nor tensorrt_lean
also seem like in the future (not now) with tensorrt v9 u can do pip install tensorrt on windows
install packaging with pip install packaging
install torch2trt with pip install git+https://github.com/NVIDIA-AI-IOT/torch2trt.git
no need to git clone
just stumble upon your repo when trying to use real-esrgan with tensorrt, i think the instructions can be a bit better, especially windows and the
.bat
filenvidia docs are very long and complicated, i write my own guide to install CUDA + cuDNN + TensorRT on windows: https://github.com/phineas-pta/NVIDIA-win/blob/main/NVIDIA-win.md (u can use this to make tutorial video if u want)
for python packages here my suggestions:
install torch if cuda 11.8:
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
if cuda 12.1:pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121
install
tensorrt
if linux:pip install tensorrt
if windows:pip install
+ the tensorrt.whl
file in the folder downloaded when install TensorRT no need to installonnx_graphsurgeon
nortensorrt_dispatch
nortensorrt_lean
also seem like in the future (not now) with tensorrt v9 u can dopip install tensorrt
on windowsinstall
packaging
withpip install packaging
install
torch2trt
withpip install git+https://github.com/NVIDIA-AI-IOT/torch2trt.git
no need to git clone