The template for running Python applications in the on the Remote run system.
When submitting a Python request the current context of the executable will be packaged and sent to the server. When on the server the requirements.txt will be installed and the run-file (set in the config) will be run with the run prams (also set in the config). Because of the "why bother" approach python has to versioning and compatibility, a general solution wold be impractical, thats why you can set the docker image to run from. The default image is nvidia/cuda:10.1-runtime . You can put stuff in the requirements.txt to be installed on top of this. To avoid having to wait for installation, a image with the packages preinstalled cold be used e.g. tensorflow/tensorflow:latest-gpu if you need a spesific verson of tf you can find an image with the matching verson here for Tf 1 you can use somthing like this tensorflow/tensorflow:1.15.2-gpu
To run successfully you need to:
- have a requirements.txt in the executable context with ALL dependencies. That is, no local dependencies, if any are needed they have to be in the local context
- Everything you want returned after the run is complete has to be saved to the the ./save_data dir
- Python 3 is used
Execute the executable called Remote-run and follow the instructions on screen. If all the requirements above are met you can run whatever you want.