Port of PYTTI notebook geared towards local execution. Integrated with ClearML for job logging, monitoring, and comparison.
Contributions welcome.
- docker
- ClearML
- either a local instance (requires docker-compose) or create a free account on the hosted service
- Will change this to not be a requirement soon.
- Also planning to add support for WandB, etc.
-
Clone and CD into this project
-
Add your
clearml.conf
to the root directory -
Build the container
$ docker build -t pytti:test .
This automates installing packages for PYTTI and downloading pre-trained models.
-
Start the container
$ mkdir /opt/colab/images_out
$ docker run --rm -it -p 8181:8181 --gpus all -v /opt/colab/images_out:/opt/colab/images_out pytti:test
You should know have a jupyter server running at http://localhost:8181/lab?token=UniqueNewYork . (You should change that token for security)
You should now be able to run the pytti beta-p5 notebook.
Additionally, the container contains a modified version of the notebook code which can be run as a script. This script is configured with OmegaConf/Hydra yaml files. To use this script:
-
Define a new experiment by adding a config file to the ./config/conf directory. You only need to specify anything you want changed from the defaults, which are specified in ./config/default.yaml . Let's say you named your experiment configuration: ./config/conf/demo.yaml
-
Open a terminal on the jupyter server
-
Run the script, passing the experiment defining config as an argument for hydra.
$ python pytti_cli_w_clearml.py conf=demo
Because the config is managed by hydra, you can override experiment parameters by specifspecified them on the command line.