Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
for context: at $8/hr (say, GB300 in an extreme supply crunch, or Vera Rubin), $100B pretrain would mean half a year on 3 million GPUs that's many orders of magnitude and solving technical impossibilities away I don't feel concerned if it'll take 1 extra year
«Pre-trains will cost hundreds of billions!»
Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.