🚀 Feature

Currently Aim supports storing artifacts in S3 and in a filesystem. Support for other clouds would be nice -- here I am asking to support Azure Blob Storage.

Motivation

My motivation - we are considering using Azure for model training. Specifically we may be running trainings on multiple clouds to optimize costs.

AIM's motivation - supporting all major providers (AWS, GCP, Azure) would increase adoption.

Pitch

See "Motivation" above.

Alternatives

Possibly this could be somehow hacked together by attempting to mount Azure Blob Storage as a filesystem. Alternatively we could send over the data from compute instances in Azure to directly to S3, but this would incur egress cost which in our case is non negligible.

Additional context

We are currently not using AIM and we are exploring our options. This feature would definitely help us for the case described above, once we get there.

We currently don't have this issue, however we are looking at different options for experiment tracking. At the very least I would like to get an understanding whether you are open to supporting this functionality. We are happy to spend some development to implement this feature when we need it, given proper support from the AIM team.