Dask backend is based on the Dask library. You can use dask on the single machine and on the cluster thus you can simply start with the simple solution and then scale it. Additional dask advantage is that after featching historical features you can continue with the distributed training.

To configure the polars offline store edit feature_store.yaml

project: feature_repo
registry: data/registry.db
provider: local
    type: yummy.YummyOfflineStore
    backend: dask