My use case is for interpretability work, running LMs locally for more control or renting GPUs to conduct this work as well as for development.
I know apple silicon has MPS that supports the fundamentals (transformerlens, nnsight) etc so 64-128gb variants of MacBooks have appealed to me. But I’m not quite sure about it yet.
What do you guys suggest? This post is not exclusive to this particular use case entirely but more so what an experienced researcher at the moment would buy depending on specific workloads
Thanks!
[link] [comments]