TAO classification /bin/sh: 1: pip3: not found

Ah.
How do I find my entrypoint, please? (I have never specified an entrypoint because I don’t know where I would find that - apologies)

I mean you need to type the correct "" instead of “” .

“docker run” presumably needs an entrypoint to .docker?

which is at /home/peter?

but apparently not

No. Please run as below.
$ docker run --runtime=nvidia -it --entrypoint="" -p 8888:8888 --rm -v /var/run/docker.sock:/var/run/docker.sock -v yourlocal_folder_where_store_notebook:/test
nvcr.io/nvidia/tao/tao-toolkit-tf:v3.22.05-tf1.15.5-py3 /bin/bash

Despite the fact that I am now simply copy/past’ing

docker run --runtime=nvidia -it --entrypoint="" -p 8888:8888 --rm -v /var/run/docker.sock:/var/run/docker.sock -v yourlocal_folder_where_store_notebook:/test
nvcr.io/nvidia/tao/tao-toolkit-tf:v3.22.05-tf1.15.5-py3 /bin/bash

and replacing “” with the absolute path to the directory containing “config.json”, because an earlier error message seemed to be looking for that.

I am currently having no success.

I am not typing the correct “” instead of “”, but where do I find more on this, it doesn’t seem to be covered in the nVidia documentation.

What is the config.json and what did you change? Could you share the config.json?

And adding the --entrypoint="" is the workaround when user run “docker run” instead of tao launcher. (See the background in Chmod: cannot access '/opt/ngccli/ngc': No such file or directory - #2 by Morganh )
Just need to set --entrypoint="" . Not needed to set any others.

I may have found the problem.

at /home/peter/.docker/config.json it looks like this:

just an authorisation code/token

at /home/peter/TAO_toolkit/config/files/config.json it looks like this:

I created this from the instructions “Creating an Experiment Spec File” found at TAO Toolkit — TAO Toolkit 3.22.05 documentation. It contains a placeholder spec. for a real-world use case.

This feels like I have been following two different sets of instructions and I need to reconcile them.

Please advise in easy steps what I need to do: I will not make any changes until I get feedback from you…

Do you mean Image Classification (TF1) - NVIDIA Docs ?

Yes

In TAO classification user guide, I do not find any info about creating config.json.
Actually it is a training spec file according to your screenshot.

In TAO, we only need to set ~/.tao_mounts.json. It is mentioned in tao launcher section. TAO Toolkit Launcher — TAO Toolkit 3.22.05 documentation

Understood.
So I have already sent you the only config.json that still exists, but here it is again:

Please advise

So, can you run below successfully now?
$ docker run --runtime=nvidia -it --entrypoint="" -p 8888:8888 --rm -v /var/run/docker.sock:/var/run/docker.sock -v yourlocal_folder_where_store_notebook:/test
nvcr.io/nvidia/tao/tao-toolkit-tf:v3.22.05-tf1.15.5-py3 /bin/bash

Hi again
I have read’ Chmod: cannot access ‘/opt/ngccli/ngc’: No such file or directory - #2 by Morganh

I have followed the approach 2

However there is an error in the path:
’ lib/python3.6/site-packages/tao/components/docker_handler/docker_handler.py’
on my machine at least it should read
lib/python3.6/site-packages/tlt/components/docker_handler/docker_handler.py

approach 2 was simple to implement (after error was found)
approach 1 just scrambled my brain

I am back in the notebook
The notebook cannot find pip3

62 exchanges later can we pick it up here?
How do I install pip3 in the notebook?
please, thank you

No, you need not use the workaround(approach 2) now. We already provide the fix. If you run tao launcher instead of “docker run”, just need to update the wheel.
$ pip3 install nvidia-tao==0.1.24

‘Requirement already satisfied’
tao list shows:

How do I install or import or download pip3 into the Jupyter notebook?

I still suggest you to figure out why you cannot find pip3 in notebook. Actually other users do not have this issue.

See below screenshot you shared previously.

Can you activate the conda environment that you have just created.

$ conda activate launcher

then,
$ pip3 install nvidia-tao

then, trigger notebook
$ jupyter notebook --ip 0.0.0.0 --allow-root --port 8888



answer “no”.
The shell environment in Jupyter notebook is supposed to match the Python version used to launch the notebook. I am paraphrasing from :
https://jakevdp.github.io/blog/2017/12/05/installing-python-packages-from-jupyter/#:~:text=%23%20Install%20a%20pip%20package%20in%20the%20current,install%20numpy%20Requirement%20already%20satisfied%3A%20numpy%20in%20%2FUsers%2Fjakevdp%2Fanaconda%2Flib%2Fpython3.6%2Fsite-packages
This is not happening here. I have precisely followed your instructions.
also it looks like the notebook is currently using python 3.7, not python 3.6

Please advise

Above is my log. May I know what’s your info?


Currently it is this.

$ which jupyter