pith. sign in

When attention sink emerges in language models: An empirical view

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it

fields

cs.CL 1

years

2026 1

verdicts

CONDITIONAL 1

clear filters

representative citing papers

Measuring Maximum Activations in Open Large Language Models

cs.CL · 2026-05-15 · conditional · novelty 7.0

A unified measurement pipeline on 27 LLM checkpoints shows activation maxima spanning four orders of magnitude, with MoE models 14-23x lower than matched dense models and residual streams carrying the global max in most cases.

citing papers explorer

Showing 1 of 1 citing paper after filters.

  • Measuring Maximum Activations in Open Large Language Models cs.CL · 2026-05-15 · conditional · none · ref 15

    A unified measurement pipeline on 27 LLM checkpoints shows activation maxima spanning four orders of magnitude, with MoE models 14-23x lower than matched dense models and residual streams carrying the global max in most cases.