Fetching the paper…

DISTFLASHATTN: Distributed Memory-efficient Attention for Long-context LLMs Training · Around