Fireside chat QA with Yuxiang Wei from Meta FAIR with S2S

Questions surrounding distributed training at FAIR

Well today I woke up at 5am to go on-site at my startup in Paris (I live 500km away, in Bordeaux) and got home at around 7pm, so not much free time to study, so I just read a bit in the train and ML related stuff was just the evening fireside chat.

Fireside chat with Yuxiang Wei from Meta FAIR (S2S)

As part of the course (conference) Scratch To Scale, we have a lot of very qualified engineers and researchers from top labs / companies dedicating time for guest lectures or QA sessions.

Today we had the chance to listen to Yuxiang Wei — of Meta FAIR, of the SWE-RL paper — and ask him loads of questions. We discussed topics such as {pre, mid, post}-training, training infrastructure, the advent of RL and its impacts on said infrastructure and more.