From 63911498e88993120639871fa9399c0fc48e2ff2 Mon Sep 17 00:00:00 2001
From: alainnothere <164234422+alainnothere@users.noreply.github.com>
Date: Wed, 18 Mar 2026 20:55:21 +0000
Subject: [PATCH] Initial commit

---
 README.md | 2 ++
 1 file changed, 2 insertions(+)
 create mode 100644 README.md

diff --git a/README.md b/README.md
new file mode 100644
index 0000000..21891db
--- /dev/null
+++ b/README.md
@@ -0,0 +1,2 @@
+# llm-circuit-finder
+I replicated Ng's RYS method and found that duplicating 3 specific layers in Qwen2.5-32B boosts reasoning by 17% and duplicating layers 12-14 in Devstral-24B improves logical deduction from 0.22→0.76 on BBH — no training, no weight changes, just routing hidden states through the same circuit twice. Tools included. Two AMD GPUs, one evening.