Shrikar Archak from Instacart talk on Fine Tuning
Details
Shrikar Archak from Instacart will talk about his fine tuning project using an insurance Question/Answer dataset where he reduced error rates by 2/3 for $.65 on a GPU.
Very impressive. I can't get a GPU to power up for $.65.
Talk from an industry ML expert deploying models to meet business goals.
Bring questions on FT architectures, how does RL perform in wild, how do you know if the trajectories are getting longer, when to stop training, and perhaps what he thinks of the current job market for sw engineers!