This Friday on Arxiv Dives, we have guest host @danieljvaroli diving into Attention Sinks and how they allow for stable language modeling and generation up to 4 million tokens. Sign up here 👉 http://oxen.ai/community