dettmers2022-llmint8
arXiv: 2208.07339
TLDR (English)
Reveals "emergent outliers" in large model activations and proposes mixed-precision solution. Core work behind bitsandbytes library, first enabling 175B models to fit in 8 A100s.
TLDR(中文)
揭示大模型激活中的"emergent outliers",并提出混合精度方案。bitsandbytes 库背后的核心工作,让 175B 模型第一次能塞进 8 卡 A100。