speech-enhancement VS AuTuMN

I find writing smoke tests (described here, code examples: model tests + training loop tests) and running them on every commit in CI (eg GitHub Actions) catches a lot of problem. Using pydantic is good for keeping any config files valid is good - you can smoke test those as well.

Even if detailed unit testing is hard, you can smoke test your models in CI to make sure that they're at least not crashing. More on smoke tests here. Some example smoke tests for a neural net here. Running your tests in GitHub Actions is relatively easy (here).

use Packer to pre-build an Amazon Machine Image (AMI) to pre-install all the garbage you need to run your code (eg NVIDIA stuff)

You can roll your own boto3 wrapper and do some pretty complicated stuff that way: https://github.com/monash-emu/AuTuMN/blob/master/autumn/remote/aws/cli.py

Buildkite can do this - example task with UI - pipeline config

The team at the Epidemiological Modelling Unit, School of Public Health at Monash Uni, Australia are building and maintaining this library for infectious disease modelling. It's currently being used to model COVID-19 and tuberculosis (see here).

Not ML but disease modelling, pretty similar project setup imo: here

It's configured via the web UI or YAML files in GitHub. I've been using it a lot this year and I really like how simple it is to use. I have had very, very bad experiences with all the other CI systems for this purpose. I also wrote a little Python DSL for defining pipelines.

use boto3 + click to create a CLI for starting, stopping, SSHing into machines: example CLI module

We're also hiring a software developer to work in the same team (job listing) to help us work on the autumn codebase, more details in the "position summary" section of the PD (here).