* Further reduce the number of alls to head for cached models/tokenizers/pipelines * Fix tests * Address review comments
RepoNotFoundError