13 Sep
2024
13 Sep
'24
11:26 a.m.
8/24 there's some energy around the transformer-curved-object-conversion (maybe a gear unsure). hard to figure out, i'm not a mechanical enginee--. i have gpt2 open locally, maybe i can look inside it. ; gpt2 ummm has umm a more complex attention kernel than newer models because trends on how to do it hadn't settled as much maybe i'll look at llama instead i dunno maybe i'll just think about it! who knows !!!!!!!