3 ms·Processing 24T tokens for LLM training with 0 crashes (what made it possible)1 points by DISCURSIVE 1y ago