I am using the TAO toolkit V5 and I see that there are a lot of environment variables that are used inside the notebooks, but most of them are not clear.
For instance, what is the different uses of these variables:
LOCAL_PROJECT_DIR, LOCAL_EXPERIMENT_DIR, USER_EXPERIMENT_DIR.
I am aware that these variables are used to run cleaner commands, however, I don’t understand the intutition behind having these three different variables that are, for me, if I understood them well, serving the same purpose.
The same goes for: LOCAL_SPECS_DIR and SPECS_DIR. Why should there be two different directories?
Because there are local folders and docker’s folders.
We need to differentiate them. Actually you can find the different paths when you set ~/.tao_mounts.json file.
One is the path to your local files.
Another is the mapping path to the docker’s files.
! tao ssd run ls /workspace/xxx.txt
/workspace/xxx.txt is a path which is inside the docker. It is not a path locally.
I see. Thanks. Much clearer now.
By the way, when I run some help command, I expect to have an imminent result on how to use it, but seems that a docker image is pulled just for that. For example, running tao model faster_rcnn --help, I get the below processing …
When I run the help command again, I should wait a couple of minutes again to get the result. Is there a workaround to avoid this?
docker pull nvcr.io/nvidia/tao/tao-toolkit:5.0.0-tf1.15.5 firstly.
Still getting the same waiting time …
You can also run inside the docker directly.
$ tao model faster_rcnn
Then inside the docker
# faster_rcnn --help
It seems I still have the same waiting time.
I went inside the docker and ran the help command from there, but there is still a waiting time. I am wondering whether this is something that all users are experiencing or only myself.
If it’s ‘natural’ that this happens, this this is ok.
From the screenshot, it cost about 3 seconds to prompt the result for “faster_rcnn --help” . It is normal.