🐛 [Bug] Encountered bug when using Torch-TensorRT

[dscho15]

Bug Description

I can't install TensorRT using pip install on Ubuntu 20.04 in a virtual environment.

To Reproduce

Steps to reproduce the behavior:

1. Install tensorrt, etc. https://pytorch.org/TensorRT/getting_started/installation.html#installation
2. pip3 install nvidia-pyindex
3. pip3 install nvidia-tensorrt
4. pip3 install torch-tensorrt -f https://github.com/pytorch/TensorRT/releases

pip install torch-tensorrt -f https://github.com/pytorch/tensorrt/releases
Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com
Looking in links: https://github.com/pytorch/tensorrt/releases
Collecting torch-tensorrt
  Downloading torch-tensorrt-0.0.0.post1.tar.gz (9.0 kB)
  Preparing metadata (setup.py) ... error
  error: subprocess-exited-with-error
  
  × python setup.py egg_info did not run successfully.
  │ exit code: 1
  ╰─> [13 lines of output]
      Traceback (most recent call last):
        File "<string>", line 2, in <module>
        File "<pip-setuptools-caller>", line 34, in <module>
        File "/tmp/pip-install-pd1uq46e/torch-tensorrt_c4b37a4bd4a440869a54423c227a1651/setup.py", line 125, in <module>
          raise RuntimeError(open("ERROR.txt", "r").read())
      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
      ###########################################################################################
      
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

× Encountered error while generating package metadata.
╰─> See above for output.

note: This is an issue with the package mentioned above, not pip.
hint: See above for details.

Expected behavior

I would not expect it to throw an error since I follow the instructions, and I have installed TensorRT correctly, since I can convert models from torch to tensorrt using MMLabs interface.

Environment

• Torch-TensorRT Version (e.g. 1.0.0):
• PyTorch Version: 1.12.1
• OS (e.g., Linux): Ubuntu 20.04
• How you installed PyTorch (conda, pip, libtorch, source): conda
• Build command you used (if compiling from source): Used .tar installation guide when I installed TENSORRT and CUDANN.
• Python version: 3.9
• CUDA version: 11.7
• GPU models and configuration: Nvidia RTX 3070

Additional context

Pip packages:
Package                   Version
------------------------- ----------
brotlipy                  0.7.0
certifi                   2022.9.14
cffi                      1.15.1
charset-normalizer        2.1.1
cryptography              37.0.4
idna                      3.4
mkl-fft                   1.3.1
mkl-random                1.2.2
mkl-service               2.4.0
numpy                     1.23.1
nvidia-cublas-cu11        2022.4.8
nvidia-cublas-cu117       11.10.1.25
nvidia-cuda-runtime-cu11  2022.4.25
nvidia-cuda-runtime-cu117 11.7.60
nvidia-cudnn-cu11         2022.5.19
nvidia-cudnn-cu116        8.4.0.27
nvidia-pyindex            1.0.9
nvidia-tensorrt           8.4.3.1
Pillow                    9.2.0
pip                       22.2.2
pycparser                 2.21
pyOpenSSL                 22.0.0
PySocks                   1.7.1
requests                  2.28.1
setuptools                65.3.0
six                       1.16.0
torch                     1.12.1
torchaudio                0.12.1
torchvision               0.13.1
typing_extensions         4.3.0
urllib3                   1.26.11
wheel                     0.37.1

[franzpeter124]

Having the same issue! Also followed https://github.com/pytorch/TensorRT/issues/1026#issuecomment-1119561746 and tried explicitly the combination torch==1.10.0 torch-tensorrt==1.0.0.

For pip install torch-tensorrt -f https://github.com/NVIDIA/Torch-TensorRT/releases I get the same error message from above (https://github.com/pytorch/TensorRT/issues/1371#issue-1383487444).

With specified version 1.0.0:

pip install torch-tensorrt==1.0.0 -f https://github.com/NVIDIA/Torch-TensorRT/releases
Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com
Looking in links: https://github.com/NVIDIA/Torch-TensorRT/releases
ERROR: Could not find a version that satisfies the requirement torch-tensorrt==1.0.0 (from versions: 0.0.0, 0.0.0.post1)
ERROR: No matching distribution found for torch-tensorrt==1.0.0

[amarckal]

It seems github is not returning a response with assets expanded, and here is a workaround

pip install torch-tensorrt==1.1.0 --find-links https://github.com/pytorch/TensorRT/releases/expanded_assets/v1.1.0

Unfortunately, for other versions you have to modify both version tags in the command above.

[dscho15]

Thanks, I'm closing this issue as it is resolved.

[amarckal]

@dscho15 I'm not a maintainer of this repo but I think you should leave this issue open, since it needs to be properly resolved i.e. update the installation instructions in the readme.

[narendasan]

Yeah seems like Github has changed the link structure for the release page and links are not directly accessible anymore without going to the specific stub for a release's artifacts. We are looking into the requirements to setup and maintain a index but in the meantime documentation has been updated