Open sebastian-sz opened 3 years ago
Feature request
Given we can have CUDA and cuDNN, it would be amazing to have TensorRT preinstalled in Google Colab. Link
Describe the current behavior:
Currently, tensorRT is unavailble in Colab:
import tensorrt as trt ModuleNotFoundError Traceback (most recent call last) <ipython-input-9-0a87b2689341> in <module>() ----> 1 import tensorrt as trt ModuleNotFoundError: No module named 'tensorrt'
in order for the above to work, one needs to undergo a tedious, and rather long process of manually installing tensorRT in Colab, which not only wastes developer time but also wastes GPU time (GPU is necessary for tensorRT installation to work).
Describe the expected behavior:
It would be amazing if tensorRT came preinstalled by default, and we could straight away do:
import tensorrt as trt
I wouldn't mind this beeing only-pro or only-gpu option.
The web browser you are using (Chrome, Firefox, Safari, etc.):
Google Chrome.
Link (not screenshot!) to a minimal, public, self-contained notebook that
reproduces this issue (click the Share button, then Get Shareable Link): N/A
Hello @sebastian-sz, you are absolutely right ... I've been trying to install this and still can't import tensorrt as trt correctly, always giving me some type of error ... Have you found a way to correctly import this? Thank you
@rafaelbate I used to do this very ugly and hacky approach (hence this Feature Request), where I copied the .deb
file into Colab and installed necessary packages via apt-get
. This approach doesn't work anymore.
yes i really need this feature please
Did anyone try the upstream instructions (https://docs.nvidia.com/deeplearning/tensorrt/quick-start-guide/index.html#install)? Where did things go wrong?
I approve of the suggestion. I tried to install TensorRT myself for a few days, but I seemingly am not able to install it correctly and I have no idea why. I tried multiple version, I recompiled various things, but I see warnings.
[01/30/2022-18:13:44] [TRT] [W] TensorRT was linked against cuDNN 8.2.1 but loaded cuDNN 8.0.5
The furthest I got was to being able to run code with warning messages, but I want to run code without getting warning messages. Most likely due to piping of data into ffmpeg, my code does not work if any text gets printed. It would be nice if properly installed TensorRT would be available by default.
Is there any update regarding this issue?
If you're struggling with this - Nvidia recently made it possible to install in Colab via pip
(this takes couple of minutes though).
!pip install --upgrade --index-url https://pypi.ngc.nvidia.com nvidia-tensorrt
And in Colab:
import tensorrt as trt
trt.__version__
# 8.4.0.6
Feature request
Given we can have CUDA and cuDNN, it would be amazing to have TensorRT preinstalled in Google Colab. Link
Describe the current behavior:
Currently, tensorRT is unavailble in Colab:
in order for the above to work, one needs to undergo a tedious, and rather long process of manually installing tensorRT in Colab, which not only wastes developer time but also wastes GPU time (GPU is necessary for tensorRT installation to work).
Describe the expected behavior:
It would be amazing if tensorRT came preinstalled by default, and we could straight away do:
I wouldn't mind this beeing only-pro or only-gpu option.
The web browser you are using (Chrome, Firefox, Safari, etc.):
Google Chrome.
Link (not screenshot!) to a minimal, public, self-contained notebook that
reproduces this issue (click the Share button, then Get Shareable Link): N/A