This repository can be private, opposite to being a fork of the course material. And it is lighter to work with. It is intended to include all the work and handins through the course.
The hot-dog / not-hot-dog app is in the app directory.
To do:
- Make the model predict correctly after some training. Just debug the training and make sure it works
- Split two copies of the model to threads and communicate rather than run on one thread? EDIT: Kinda started on the
threadingbranch here. - Make the images a part of repo.
- Make the model path not absolute.
- The overview of the architecture and essays on JAX (2pt) and Model Distributed Parallelism (1pt) are here.
The code is here and the associated text (just some learnings) are here.