V0.1.0

Overview of Changes

Consists of PRs: #119, #116, #115, #113, #108

• Rafiki’s data (i.e. model, job, user data) now persists across restarts
• Fixed “out-of-memory” error when large models are trained on Rafiki (by moving storage of model parameters from database to disk)
• App developers now can specify no. of GPUs to use for a train job, and during model training, each model is exclusively allocated a single GPU, if GPU is available
• Admins can now ban users
• Model developers can now delete models
• Stricter authentication checks on the back-end for management of models, users and jobs
• Better documentation (deployed at https://nginyc.github.io/rafiki/docs/0.1.0)
• User tokens now expire in 1h