Want to create an interactive transcript for this episode?
Podcast: Arxiv Papers
Episode: Active-Dormant Attention Heads: Mechanistically Demystifying Extreme-Token Phenomena in LLMs
Description:
The paper investigates extreme-token phenomena in transformer-based LLMs, revealing mechanisms behind attention sinks and proposing strategies to mitigate their impact during pretraining.
https://arxiv.org/abs//2410.13835
YouTube: https://www.youtube.com/@ArxivPapers
TikTok: https://www.tiktok.com/@arxiv_papers
Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016
Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers