Uber's ML Systems (Uber Eats, Customer Support), Declarative Machine Learning - Piero Molino - The Data Scientist Show #064
The Data Scientist Show - Daliana Liu
00:00
The Challenges of Productionizing Large Language Models
DBT co-pilot: It's easy to build a demo, but it's hard to put things in production. Security and the fact that some data cannot leave their clouds or their premises is one aspect. The other aspect is even when it can leave the premises, whenever you get to a point where the amount of interactions, among API calls, that you need to do passes a certain threshold, you're too expensive. And finally, if you have above a certain amount of data that you're actually working with, for instance, streaming cases, in those cases, it becomes invisible to wait for the model answer,. So it's just the current MLM APIs are just too slow.
Transcript
Play full episode
