---
license: gemma
base_model: google/gemma-2-2b-it
tags:
  - rys
  - layer-duplication
  - reasoning-circuits
  - gguf
  - sovereign-collection-v2
---

# Gemma-2-2B-RYS-19-23

Gemma-2-2B-it with layers 19-22 duplicated. A late-stack reasoning circuit runs twice on every forward pass.

26 base layers → 30 after duplication. No training, no merging, no weight changes.

**Reasoning 58.82% → 64.70% (+5.88). EQ 91.72 → 93.87 (+2.15). Math 0.522 → 0.5027 (−1.93).** Peak reasoning Δ across the full 36-config sweep is +17.65%; (19,23) is the best-combined pick.

## Results

| Metric | Baseline | RYS (19,23) | Delta |
|--------|----------|-------------|-------|
| Math | 0.522 | 0.5027 | −1.93 |
| EQ | 91.72 | 93.87 | +2.15 |
| Reasoning | 58.82% | 64.70% | +5.88 |

**The smaller-Gemma.** Within-family control for the larger Gemma-2-9B sibling. The two share baseline reasoning (58.82%) but the 9B finds richer duplicate-circuit candidates (peak +23.53% with 17 boosters) than this 2B (peak +17.65% with 4 boosters). Depth-room scaling: at matched baseline, more layers means more candidate blocks where duplication moves the reasoning probe.

Pick this when you want a tiny Gemma with reasoning lifted modestly and EQ slightly improved.

## Usage

```
llama-server -m Gemma-2-2B-RYS-19-23-Q4_K_M.gguf -ngl 99
```

## Full sweep data

36 configurations tested. (19,23) block-4 is the best-combined pick. Full per-config sweep + cross-architecture analysis: [v2 dataset](https://huggingface.co/datasets/john-broadway/rys-sovereign-collection-v2).


Part of the RYS Sovereign Collection v2.

---

## Where this sits in the Sovereign Collection

**v1 — Qwen2.5 cross-scale + Qwen3-32B headline crossover.** 5 model repos.

**v2 — cross-architecture corpus.** 21 model variants across 10 architecture families. Inverse correlation (r = −0.726): weak baselines lift more, in their weakest dimension. 13 deployable RYS-applied weight repos covering every non-zero-lift variant.

**Within-family sibling:** [`john-broadway/Gemma-2-9B-RYS-14-18-GGUF`](https://huggingface.co/john-broadway/Gemma-2-9B-RYS-14-18-GGUF) — the bigger Gemma with the same baseline and more lift, plus the highest baseline EQ in the corpus.

**Credit**

John Broadway, with collaboration from Claude (Opus 4.6 in April 2026 sweep generation and build pipeline; Opus 4.7 in May 2026 cross-architecture analysis and publication). Original RYS method by [David Ng](https://dnhkng.github.io/posts/rys/) on Qwen2-72B; sweep + probe toolkit by [alainnothere](https://github.com/alainnothere/llm-circuit-finder).