To support the ML-Team's MVP for model serving, we'd need a swift account to store models on that supports the S3 API. After a chat with @fgiunchedi, we thought to:
- Use Swift Thanos as interim solution for the MVP.
- Move ML Models to Swift MOSS when it will be ready.
If this is ok for everybody, we'd need an account for Swift Thanos to store the initial models. Long term we'll probably want a read-only account to use in Kubeflow, and a read/write to push models (still unclear how it will be done).