How to specify which TAO version to download when running from docker container

[ahaselhan]

Please provide the following information when requesting support.

• Hardware (4060Ti)

I am currently running TAO in a docker container as per this post.

I am at the point in the Jupyter notebook where I install TAO. Can I specify which version of TAO I want to download?

I’ve tried adding --version 5.5.0 to the command when I install, which seems to make it look for a version, but it cannot find version 5.5.0. What would be the correct command to use.

Thanks

[Morganh]

@ahaselhan

Can I specify which version of TAO I want to download

When you run with docker run instead of tao-launcher, you just need to specify the TAO version by changing the docker name(for example, nvcr.io/nvidia/tao/tao-toolkit:5.5.0-pyt) . Other versions or branches can be found in TAO Toolkit | NVIDIA NGC.
Inside the docker, it is not needed to install tao-launcher. You can ignore tao-launcher if you run with docker run.

[ahaselhan]

I’ll outline my steps, so maybe you can point out where I am going wrong. Here are the exact steps I followed:

$ docker run --runtime=nvidia -it --rm --ipc=host --gpus all --name versiontest -d -v /localhome/local-xxx:/localhome/local-xxx -p 8888:8888 -v /var/run/docker.sock:/var/run/docker.sock -v /home/porter/versiontest:/home/porter/versiontest -m 1100G --oom-kill-disable --ulimit memlock=-1 nvcr.io/nvidia/tao/tao-toolkit:5.5.0-pyt

$ docker exec -it versiontest /bin/bash

apt-get update

apt-get install sudo

sudo apt-get update

sudo apt install docker.io

sudo ls -la /var/run/docker.sock

$docker login nvcr.io
Username: $oauthtoken
Password: (specifying your own password)

cd
cd /home/porter/versiontest
git clone https://github.com/NVIDIA/tao_tutorials

jupyter notebook --ip 0.0.0.0 --allow-root

I then opened http://127.0.0.1:8888 in my browser, and enter the token generated in my command line. I then get to the point in the jupyter notebook to check my TAO version, and get this output

Would you be able to point out where I am going wrong with this?

Thanks

[Morganh]

@ahaselhan

notebook to check my TAO version, and get this output

Would you be able to point out where I am going wrong with this?

Inside the docker, you can ignore the tao launcher. For example, you can run training without the tao in the beginning.
More, you can run with command line instead of cells of notebook.

[ahaselhan]

How would I be able to modify this statement to get it to run training?

I tried a few variations of this statement, but was not about to get it to run

[Morganh]

Please try this:
! classfication_tf2 train -e $SPECS_DIR/spec.yaml

You can also run in the terminal(putty, etc)

[ahaselhan]

@Morganh

! classfication_tf2 train -e $SPECS_DIR/spec.yaml

I tried that, but no success

How am I able to run from the command line? Is there instructions somewhere? I navigated to the location where I cloned the github repository, tried tao --help, but couldn’t find tao as a command.

[Morganh]

$ docker run --runtime=nvidia -it --rm --ipc=host --gpus all --name versiontest -d -v /localhome/local-xxx:/localhome/local-xxx -p 8888:8888 -v /var/run/docker.sock:/var/run/docker.sock -v /home/porter/versiontest:/home/porter/versiontest -m 1100G --oom-kill-disable --ulimit memlock=-1 nvcr.io/nvidia/tao/tao-toolkit:5.5.0-tf2

Then,
# classfication_tf2 train -e path/of/spec.yaml

Note, for classfication_tf2 network, need to use tf2 series docker, for example, 5.5 version:nvcr.io/nvidia/tao/tao-toolkit:5.5.0-tf2

[yingliu]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks.