Training & Pre-training
Months of thousands of GPUs predicting the next token, trillions of times.
Está a ver um explicador de conceito: o que este termo significa em linguagem acessível, e os registos que relaciona.
Pre-training runs a model over trillions of tokens, nudging weights after every prediction error, in synchronized runs across tens of thousands of accelerators — the workload the biggest clusters on this site were built for. A frontier run costs hundreds of millions of dollars in compute, fails in exotic ways (silent data corruption, stragglers, loss spikes), and is why checkpointing, interconnects and reliability engineering are front-page infrastructure topics.
Como este elemento se liga ao resto da stack
As ligações deste registo num relance — clique num grupo para ir diretamente até ele.
Relacionados
Fontes
Llama 3 herd paper (training at scale) ↗
Registro revisado em 2026-01-15 — ver Metodologia.