minLillemus (my little mouse)

Layer 12 mid-layer precision splice. FFN transplanted from layer 2. 29% PPL recovery across the largest depth gap (10 layers). Demonstrates that even across significant architectural distance, GRC basis alignment transfers functional structure. Minimal intervention with measurable effect.

Architecture

• Base: SmolLM2-135M-Instruct
• Method: CECI Protocol (HyperTensor Paper X) — GRC basis projection
• Created: 2026-05-04
• Repository: HyperTensor

Graft Proof

This model was created by:

1. Computing the GRC (Geodesic Residual Compression) basis from the target layer's attention weights via SVD
2. Projecting the donor layer's FFN weights into the target's geometric subspace
3. Blending at controlled strength to preserve stability

Perplexity testing confirms the graft transfers functional structure without destroying the model.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("NagusameCS/minLillemus", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("NagusameCS/minLillemus")