3 ms·
Isn't a key selling point of the latest, hottest model that's on the front page of Hacker News multiple times right now, the fact that it fits on consumer-grade
by macksd 4y ago
Isn't a key selling point of the latest, hottest model that's on the front page of Hacker News multiple times right now, the fact that it fits on consumer-grade GPUs? Surely some of the interesting ideas it's spawning right now are people doing transfer learning on GPUs that don't end in "100", don't you think?
- fragmede 4y agofor what it's worth, stable diffusion was trained on 32 x 8 x A100 GPUs
- macksd 4y agoYou know there's a huge difference between training the original model and transfer learning to apply it to a new use case, right? Saying people are years behind if they think there work is only worth something with 8 A100 pods is pretty ignorant of how most applications get built. Not everyone's trying to design novel model architectures, nor should they.