67 - GLUE: A Multi-Task Benchmark and Analysis Platform, with Sam Bowman
NLP Highlights
00:00
How to Train a Multi-Task Model
Multi-task models are what gets us our best performance. We found that adding ALMO to anything is giving you improvements in performance consistently across all of these tasks. Pre-training on some subset of these tasks will generally get you decent performance on the rest of them. And probably unsurprisingly, using attention, using large models really helps you. It's a really interesting set of experiments that you ran it sounds like I need to look at more closely.
Transcript
Play full episode
