PyTorch Models (adapted from existing code) and Dockerfile

Any (or all) of
A- PyTorch Video Diffusion model, can be a simplified/stripped down version of one of these two codebases applied to moving MNIST
1- Ideally https://github.com/voletiv/mcvd-pytorch
2- Or RaMVID https://github.com/Tobi-r9/RaMViD

B- datascience / ML Dockerfile
- Dummy variables for user/cluster info
- Can scale number of GPUs
- nvidia CUDA and cuDNN version changeable (FROM an nvidia image is fine)
- conda package management
- only torch necessary, but include lines with comments for other installations

Stripped down/simplified
Project structure should be
- Dockerfile
- setup.sh
- environment.yaml
- model_config.yaml
- other_config_if_needed.yaml
- run.py
- code
- model.py
- other_files
- ...

Notes
- All input variables must be contained in the yaml files and logically ordered
- Other dependencies ideally removed
- setup.sh contains lines needed to setup environment (including image creation)