Analysis of Metastable States in the Transformer Activation Space

·LessWrong··

Part 1: Do Metastable Token Clusters exist in Trained Transformers? This is the first entry in a sequence. Over about ten parts, this series will work through a few humble experiments that test a mathematical theory of attention against real trained transformers.A project summary: a recent paper by Geshkovski, Letrouit, Polyanskiy, and Rigollet models attention as a dynamical system on the sphere and proves that tokens cluster and drift toward consensus, with a metastable two-timescale structure...

Read full article →

Related Articles

Anthropic S-1
SiNTEx · Hacker News · 18h ago
Solar power is becoming bigger than nuclear
geox · Hacker News · 16h ago
The Exponential Time Hypothesis Is False [pdf]
vok · Hacker News · 9h ago
Oldest GUI application that still runs on Windows 11
enlyth · Hacker News · 13h ago
Type Server Protocol
adamnemecek · Hacker News · 13h ago