Content
@
0 reply
0 recast
0 reaction
Giuliano Giacaglia
@giu
Out of the 350k GPUs that Meta has, they are probably using around 300k for inference. That's a big percentage just for inference
2 replies
2 recasts
34 reactions
Paul Dowman 🔴✨
@pauldowman.eth
And 50k is still a hell of a lot for training... I read GPT-4 was 25k, and that's A100s
0 reply
0 recast
0 reaction
