TAO on Google Colab

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Hello!
Is there a currently working way to run TAO on Google Colab?
I searched here in the forum and on the internet but I didn’t find a solution for the current problem that occurs when testing with the notebooks available on Running TAO Toolkit in the Cloud

The error: /bin/bash: line 1: tao: command not found

I found a post that says it doesn’t work because the default Ubuntu version in Colab recently changed to 22.04, which doesn’t support version 3.6 via deadsnake (as it was for ubuntu 20.04).
I’m sorry if this has already been answered somewhere but I couldn’t find it, so I’d like to know if there’s any way to run it in Colab, it could be any sample notebook (it doesn’t have to be the latest ones).

By the way, the “Use fallback runtime version” (Colab command when connected to a runtime) is no longer available, is that it?

The requested information.

• Hardware (T4)
• Network Type (All options)
• How to reproduce the issue ? Colab: Google Colab

2

Hi @gabriel_av07,
I believe there is no currently working way to run TAO on Colab, but the NVIDIA team is working on it.

This is the message I got:

Please check if you meet the same error as Google Colab ssd.ipynb no python3.6 - #20 by akash.singh1 .

If yes, as mentioned in that topic, Google Colab ssd.ipynb no python3.6 - #24 by Morganh , TAO team is working on the common issue in colab.

And they say the problem is that “It is due to Ubuntu default version being 22.04 in colab now, which doesn’t support python 3.6 via deadsnakes like we do for ubuntu 20.04.
Internal team are still working on that.”

4

TAO team is working on the blocking issue in Colab. BTW, if possible, you can run in local dgpu machine or cloud(such as AWS).

5

Hi @aurelio.salton and @Morganh!

Oh I see, the error that appeared to me is not exactly the same from the topic you mentioned, @aurelio.salton, but I believe they are related and being caused by the same issue.
Anyway, I’m glad to know that NVIDIA team is already working on it

Thanks for the attention, and thank you @Morganh for clarifying.
We will then wait for the official solution.

6

Hi @Morganh
I am posting again so that this topic isn’t closed. Any news on Colab? Is NVIDIA giving up on it?

7

Unfortunately, the Ubuntu version in colab has been changed to 22.04. But fallback runtime is no more available.
For TAO, it is expected to run in Ubuntu 20.04. TAO Toolkit Quick Start Guide - NVIDIA Docs .

For temporary workaround, please try to run TAO on local dgpu machine or cloud machine.

https://docs.nvidia.com/tao/tao-toolkit/text/tao_toolkit_quick_start_guide.html#hardware-requirements
https://docs.nvidia.com/tao/tao-toolkit/text/running_in_cloud/overview.html