Evaluating Language Models
The Data Exchange with Ben Lorica
00:00
Helm: A Center for Foundation Models
Helm is trying to go as fast as we can in terms of keeping up with models. We're benchmarking the new DaVinci 003 model from OpenAI. Another class of things we're training is models which are sort of unique in some way. And there's also systems level improvements like flash attention that make training go two to three times faster.
Transcript
Play full episode
