07Explain mixed-precision training: FP16 vs BF16, loss scaling, and where the numerics break.▼hard★ EssentialNVIDIAOpenAIGoogle2 repliesunlockedMixed precision is standard at scale, and the interviewer wants the numerics: why FP16 needs loss scaling, why BF16 mostly does not, and what stays in FP32. The signal is understanding dynamic range vs precision. Here is that answer.Open full answer →
63How does FP8 training work on Hopper GPUs, and how do you keep it numerically stable?▼expertNVIDIAOpenAIMistral1 replies◆ premiumFP8 promises to roughly double training throughput over BF16, but with only a handful of mantissa bits the numerics are unforgiving. Per-tensor scaling and a selective recipe are what make it converge.Open full answer →
