Luoshu Kit V0.2 CNN -> ResNet

LuoshuKit is a plug-in. Inject it into any model.

A mechanistic interpretability layer — every value now has a computed address instead of being searched.

LuoshuKit implements a structured addressing layer for neural representations, where internal values are assigned addresses that can be directly decoded rather than located through search.

---

ResNet Structural Signal

This repository abstracts the addressing mechanism beyond the original Luoshu structure, forming a generalized addressing layer.

Tested on ResNet (layer2) as a representative convolutional backbone, using a structured addressing mechanism derived from Luoshu-like principles.

For the full Luoshu addressing system (V0.3), see the Transformer-based implementation: https://github.com/luolearning/luoshu-kit-transformers

ResNet layer2: A0 → A1 → A2

• A0: no structure
• A1: anchor only
• A2: path activated

---

Early Signal

---

Usage

Install:

git clone https://github.com/luolearning/luoshu_kit.git
cd luoshu_kit
pip install -e .

Minimal example (python)

from luoshu_kit.luoshu_kit_a2_proto import inject
bridge = inject(model, layer_name="layer2")