@amitkumarvish My only solution so far has been to increase embedding model capacity on Azure ... still comparably slow.