Instructions to use john-broadway/Llama-3.2-1B-RYS-10-13-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use john-broadway/Llama-3.2-1B-RYS-10-13-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf john-broadway/Llama-3.2-1B-RYS-10-13-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/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf john-broadway/Llama-3.2-1B-RYS-10-13-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/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M

Use Docker

docker model run hf.co/john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use john-broadway/Llama-3.2-1B-RYS-10-13-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "john-broadway/Llama-3.2-1B-RYS-10-13-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "john-broadway/Llama-3.2-1B-RYS-10-13-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M

Ollama
How to use john-broadway/Llama-3.2-1B-RYS-10-13-GGUF with Ollama:
```
ollama run hf.co/john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M
```
Unsloth Desktop

How to use john-broadway/Llama-3.2-1B-RYS-10-13-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use john-broadway/Llama-3.2-1B-RYS-10-13-GGUF with Docker Model Runner:
```
docker model run hf.co/john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M
```

Lemonade

How to use john-broadway/Llama-3.2-1B-RYS-10-13-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Llama-3.2-1B-RYS-10-13-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use john-broadway/Llama-3.2-1B-RYS-10-13-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use john-broadway/Llama-3.2-1B-RYS-10-13-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "john-broadway/Llama-3.2-1B-RYS-10-13-GGUF:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Llama-3.2-1B-RYS-10-13-GGUF

A layer-duplication ("RYS" — Repeat Your Self, David Ng) variant of meta-llama/Llama-3.2-1B-Instruct: transformer layers 10–12 are duplicated, expanding the stack from 16 to 19 layers. No training, no merging, no weight changes — purely structural duplication. GGUF (imatrix Q-quants).

⚠️ Evaluation status — please read (updated 2026-06)

The large reasoning gain originally reported for this model was a measurement artifact, not a real capability gain. This card is being corrected to say so plainly.

The original card reported reasoning 0.00% → 64.71%. That 0% baseline came from a degraded inference setup, not from the model. On a current llama.cpp build the unmodified Llama-3.2-1B-Instruct already scores about 52.94% on the same reasoning probe, and this (10,13) duplication adds ~0 over that baseline (and slightly lowers the EQ probe). The headline "+64.71" was the old stack's broken floor rising back to normal — not something the duplication unlocked.

Why: the scores come from a lightweight search probe (16 math / 16 EQ / 17 reasoning questions, greedy-decoded) used to locate productive layer blocks — not a validated benchmark. Reasoning moves in steps of 1/17 ≈ 5.9%, so the deltas are coarse, and a degraded baseline can manufacture a huge apparent gain.

Independent benchmark (lm-eval-harness, GSM8K 5-shot, N=100): base 34% strict-match (38% flexible); RYS (10,13) 23% strict (27% flexible) — same 100 problems. So on a real benchmark the duplication does not improve reasoning and if anything lowers it. This both confirms the base is far from "0%" and refutes the "+64.71pp" claim directionally. (N=100; a larger paired run would tighten the magnitude, but the direction is clear.)

Bottom line: treat this as a normal Llama-3.2-1B-Instruct with layers 10–12 duplicated. Published for transparency and reproducibility of the RYS sweep — not as an improved reasoner.

Original sweep numbers (search probe — kept for the record)

probe	reported baseline	reported (10,13)	re-test note
Reasoning (17 q)	0.00%	64.71%	baseline was a degraded-stack artifact; correct-stack baseline ≈ 52.94%, (10,13) Δ ≈ 0. Real GSM8K (N=100): base 34%, (10,13) 23% — duplication lowers it
EQ (16 q)	27.11	90.12	baseline also stack-dependent; not a validated EQ benchmark
Math (16 q)	0.536	0.711	search-probe score, unconfirmed

Run it

llama-server -m Llama-3.2-1B-RYS-10-13-Q4_K_M.gguf -ngl 99

Method · data · attribution

Method: layer duplication — Repeat Your Self (David Ng); toolkit llm-circuit-finder (alainnothere).
Raw sweep data: rys-sovereign-collection-v2.
Built by John Broadway with Claude. The method and the raw data are real and reproducible; the interpretation of the original probe deltas as capability is what this update corrects.

License

Llama 3.2 Community License (inherits from the base model).

Downloads last month: 8

GGUF

Model size

1B params

Architecture

llama

Hardware compatibility

4-bit

Model tree for john-broadway/Llama-3.2-1B-RYS-10-13-GGUF

Base model

meta-llama/Llama-3.2-1B-Instruct

Quantized

(418)

this model

Collection including john-broadway/Llama-3.2-1B-RYS-10-13-GGUF

RYS Sovereign Collection v2

Collection

Qwen3-family cohort (0.6B/1.7B/8B) + cross-architecture sweep corpus (21 variants, 10 families). v1 collection: Qwen2.5 cross-scale + Qwen3-32B. • 18 items • Updated May 13