For LLMs, IBM’s NorthPole chip overcomes the tradeoff between speed and efficiencyResearchPeter Hess26 Sep 2024AI HardwareExploratory ScienceGenerative AISemiconductors
Improving Hugging Face training efficiency through packing with flash attentionTechnical noteRhui Dih Lee, Arthur Zucker, Achintya Kundu, Laura Wynter, Raghu Ganti, and Mayank Mishra28 Aug 2024AI