🐛 [Bug] Impossible to install in Windows

[enislalmi]

Bug Description

There is a bug, which I believe for uncompatibility between releases that gives me this error when I try to install the package using the official documentation.
RuntimeError:
###########################################################################################
The package you are trying to install is only a placeholder project on PyPI.org repository.
To install Torch-TensorRT please run the following command:

  $ pip install torch-tensorrt -f https://github.com/NVIDIA/Torch-TensorRT/releases

To Reproduce

Steps to reproduce the behavior:

1. Install CUDA=11.8, Install TensorRT==8.6.1 and cuDNN 8.9.0 following official NVIDIA Documentation.
2. Install Pytorch v2.0.0cu118
3. run pip install torch-tensorrt -f https://github.com/NVIDIA/Torch-TensorRT/releases

Expected behavior

I expect that after this step the library should be installed

Environment

Build information about Torch-TensorRT can be found by turning on debug messages

• Torch-TensorRT Version (e.g. 1.4.0):

• PyTorch Version (e.g. 2.0.0): (also tried with 1.13)

• CPU Architecture:

• OS (e.g., Windows):

• How you installed PyTorch (conda, pip, libtorch, source): pip

• Build command you used (if compiling from source): pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118

• Are you using local sources or building from archives:

• Python version: 3.10

• CUDA version: 11.8 (when tried with pytorch 1.13, i tried with cuda11.6 as per documentation and tensorrt 8.5 as per documentation, cudnn 8.9.0)

• GPU models and configuration: NVIDIA GEFORCE RTX 3050

• Any other relevant information: Tried with Pytorch 1.13cu116. Along changed also other libraries to be compatible TensorRT 8.5 and Cuda 11.6, but the results are the same. I have the same error.

[gs-olive]

Hello - currently, there is experimental Windows support in Dynamo, for which there is a relevant issue here: #2577. The current set of step-by-step installation instructions can be found here: #2577 (comment). There have been many helpful comments from others on that post related to issues which can come up in compilation

[enislalmi]

@gs-olive Hi! At the end I was able to solve following your comment on the issue, even though I have some warnings. However, I am not able to go beyond RuntimeError: Tried to instantiate class 'tensorrt.Engine', but it does not exist! Ensure that it is registered via torch::class_, when I do trt_model_fp32 = torch_tensorrt.compile(model, inputs=[torch_tensorrt.Input((1, 3, 112, 272), dtype=torch.float32)], enabled_precisions={torch.float32}, # Run with FP32 workspace_size=1 << 22 )

[gs-olive]

Thanks for the follow-up. This is likely due to the absence of the C++ runtime in the current windows integration. You can specify use_python_runtime=True to get around this issue.