中国开源模型预训练算力仍不及LLaMA 3.1 405B的三分之一

Periodic reminder that not a single one of these awesome Chinese open models – NOT! ONE! – had used even a third of LLaMA 3.1 405B's pretraining compute. And still don't have some of its capabilities. The East hasn't yet caught up to Meta's level from… yep, 23 months ago.