Instructions to use john-broadway/Gemma-2-9B-RYS-14-18-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use john-broadway/Gemma-2-9B-RYS-14-18-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="john-broadway/Gemma-2-9B-RYS-14-18-GGUF",
	filename="Gemma-2-9B-RYS-14-18-Q4_K_M.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use john-broadway/Gemma-2-9B-RYS-14-18-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M

Use Docker

docker model run hf.co/john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use john-broadway/Gemma-2-9B-RYS-14-18-GGUF with Ollama:
```
ollama run hf.co/john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M
```

Unsloth Studio new

How to use john-broadway/Gemma-2-9B-RYS-14-18-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for john-broadway/Gemma-2-9B-RYS-14-18-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for john-broadway/Gemma-2-9B-RYS-14-18-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for john-broadway/Gemma-2-9B-RYS-14-18-GGUF to start chatting

Docker Model Runner
How to use john-broadway/Gemma-2-9B-RYS-14-18-GGUF with Docker Model Runner:
```
docker model run hf.co/john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M
```

Lemonade

How to use john-broadway/Gemma-2-9B-RYS-14-18-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull john-broadway/Gemma-2-9B-RYS-14-18-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Gemma-2-9B-RYS-14-18-GGUF-Q4_K_M

List all available models

lemonade list

Gemma-2-9B-RYS-14-18

Gemma-2-9B-it with layers 14-17 duplicated. The early-layer reasoning circuit — at ~33% depth in a 42-layer stack — runs twice on every forward pass.

42 base layers → 46 after duplication. No training, no merging, no weight changes.

Reasoning 58.82% → 82.35% (+23.53). Math 0.539 → 0.5576 (+1.86). EQ 94.06 → 92.85 (−1.21).

Results

Metric	Baseline	RYS (14,18)	Delta
Math	0.539	0.5576	+1.86
EQ	94.06	92.85	−1.21
Reasoning	58.82%	82.35%	+23.53

The early-reasoner. Gemma-2-9B has the highest baseline EQ in the entire v2 corpus (94.06) and the earliest reasoning peak position — layer 14 of 42 (~33% depth), significantly shallower than the late-layer peaks at ~60% depth seen in Llama-3.1-8B and Mistral-7B-v0.3. Duplicating the early reasoning block lifts reasoning +23.53% while EQ stays essentially stable (−1.21). The clean reasoning gain without EQ trade-off — unlike Granite-3.1-1B-A400M's MoE-routing trade-off — suggests Gemma's instruction-tuning recipe routes reasoning through earlier, more isolatable layers.

Pick this when you want a strong reasoning lift and the highest baseline conversational fluency in the collection.

Usage

llama-server -m Gemma-2-9B-RYS-14-18-Q4_K_M.gguf -ngl 99

Full sweep data

48 configurations tested. (14,18) block-4 is the best-combined pick. Full per-config sweep + cross-architecture analysis: v2 dataset.

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. Gemma-2-9B sits slightly above the curve at its baseline — the shallow circuit position and clean EQ profile make it a notable distinct response within the corpus. 13 deployable RYS-applied weight repos covering every non-zero-lift variant.

Within-family sibling: john-broadway/Gemma-2-2B-RYS-19-23-GGUF — the matched-baseline smaller sibling that lifts less, demonstrating depth-room scaling.

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 on Qwen2-72B; sweep + probe toolkit by alainnothere.