X · @teortaxesTex
· X / Twitter
it's really, really hard to improve much on DeepSeek No matter how much compute you have, odds are that your grid searches won't find a much better gl…
it's really, really hard to improve much on DeepSeekNo matter how much compute you have, odds are that your grid searches won't find a much better global optimum. By V1, they were locked inI remember @andrew_n_carr getting a lot of use out of their hparams after V2Hiroki Naganuma: Everyone trains neural nets with learn