Ask HN: Tools for using spot instances for ML training?

I'm starting to get into training large networks. I have a fair amount of cloud credits on Azure, and I'd love to spend them efficiently and use spot instances for training.

For motivation: a regular A100 instance on Azure in US West 3 costs $2682 on demand and $291 as a spot instance.

Are there any good tools for distributed training in PyTorch on spot instances?