Attention as a Routing Graph: Live Circuit Extraction from a Single Forward Pass
arXiv:2609.25285v1 Announce Type: new Abstract: Finding circuits in language models usually means running many careful interventions. We try something simpler: treat attention as a routing map from one forward pass, keep a small set of routes that point toward the answer, and ask whether those rout…
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-09-23 04:00 · arXiv cs.AI
Attention as a Routing Graph: Live Circuit Extraction from a Single Forward Pass