Following the KServe LocalModelCache design:
- LocalModleCache is using /mnt/model/model where as GKM is using /mnt/kernel-caches. Need to understand why it is model/model and potential duplicate the pattern in GKM.
- Extract the cache in the OCI Image to a hashed subdirectory on the Node. That way if the same GKMCache is matched on multiple NodeGroups, it will only download once.
Following the KServe LocalModelCache design: