The Deep Learning DS Training Worker is available as Python package.
Installation of the Python Package (Linux and Windows)
The Deep Learning DS Training Worker is also available for Python and can be installed via the pip package manager. In order to increase the training speed, we recommend using Nvidia GPUs. For this purpose, you need to have installed CUDA® (https://developer.nvidia.com/cuda-zone). Our training worker relies on the PyTorch framework. The binaries of PyTorch are not directly distributed via PyPI for all platforms but by PyTorch. Hence, the PyTorch repository is to be specified during installation. Depending on the CUDA version, install the training worker via
# CPU pip install dlds-training[cpu] -f path/to/dlds_training[...].whl -f https://download.pytorch.org/whl/cpu/torch_stable.html # CUDA 9.2 pip install dlds-training[cuda] -f path/to/dlds_training[...].whl -f https://download.pytorch.org/whl/cu92/torch_stable.html # CUDA 10.1 pip install dlds-training[cuda] -f path/to/dlds_training[...].whl -f https://download.pytorch.org/whl/cu101/torch_stable.html # CUDA 10.2 pip install dlds-training[cuda] -f path/to/dlds_training[...].whl -f https://download.pytorch.org/whl/cu102/torch_stable.html # CUDA 11.0 pip install dlds-training[cuda] -f path/to/dlds_training[...].whl -f https://download.pytorch.org/whl/cu110/torch_stable.html
The Deep Learning DS Client will be installed as well and offers the interface to the worker.
Once you have installed the Training Worker, it can be started via the command
dlds-worker. It is recommended to configure the worker to run in the background. On Linux, you can make use of a tool such as supervisor for this purpose.
As soon as your worker is running, you can set it up using command-line with the Deep Learning DS Client or if you favor a graphical interface, you can use the Deep Learning DS Training Hub (visit the Section “Releases“ below).
dlds worker setup and follow the instructions to set up the worker. You will be prompted for credentials for Deep Learning DS, the hostname of the Deep Learning DS Server (defaults to api.vision.data-spree.com), and the local directory to store datasets for training. Make sure to specify the location correctly. During setup, a new worker instance will be created at the server. Once created, you can select and use the worker when starting trainings online.
Once the Training Worker is set up and running, you can get its state via the command
dlds worker status.
The Training Hub offers a small graphical user interface to interact with the worker. When starting the Training Hub, it will connect to the running worker. In case the worker is not yet set up, a form will be shown to enter the necessary credentials. Afterward, you can see the status and re-run the setup if necessary.