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.
alainnothere/llm-circuit-finder
View on GitHub ↗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.
llm-circuit-finder grew +0.0% in the last 30 days, ranks 41st of 71 in Interpretability & Alignment, last commit 219 days ago.
Star trajectory
115 days tracked+0.0% in 30d, below the Interpretability & Alignment median of +0.2%.
Last commit 219 days ago, staler than the median of 202d.
How it compares in Interpretability & Alignment
Ranked 41st of 71 by stars.
- Language
- Python
- License
- -
- Contributors
- 1
- Forks
- 20
- Watchers
- 5
- Open issues
- 2
- Releases
- 0
- Created
- 2026-03-18
- Stars this week
- +0
You've seen where llm-circuit-finder stands today.
Get the moves before everyone else: the 5 Interpretability & Alignment repos climbing fastest, every Tuesday.
Free weekly digest. Double opt-in, unsubscribe anytime.