Instructions to use john-broadway/Qwen3-32B-RYS-20-28-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use john-broadway/Qwen3-32B-RYS-20-28-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="john-broadway/Qwen3-32B-RYS-20-28-GGUF",
	filename="Qwen3-32B-RYS-20-28-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/Qwen3-32B-RYS-20-28-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/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf john-broadway/Qwen3-32B-RYS-20-28-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/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf john-broadway/Qwen3-32B-RYS-20-28-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/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf john-broadway/Qwen3-32B-RYS-20-28-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/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf john-broadway/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M

Use Docker

docker model run hf.co/john-broadway/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use john-broadway/Qwen3-32B-RYS-20-28-GGUF with Ollama:
```
ollama run hf.co/john-broadway/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M
```

Unsloth Studio new

How to use john-broadway/Qwen3-32B-RYS-20-28-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/Qwen3-32B-RYS-20-28-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/Qwen3-32B-RYS-20-28-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/Qwen3-32B-RYS-20-28-GGUF to start chatting

Pi new

How to use john-broadway/Qwen3-32B-RYS-20-28-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf john-broadway/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/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/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use john-broadway/Qwen3-32B-RYS-20-28-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf john-broadway/Qwen3-32B-RYS-20-28-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/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M

Run Hermes

hermes

Docker Model Runner
How to use john-broadway/Qwen3-32B-RYS-20-28-GGUF with Docker Model Runner:
```
docker model run hf.co/john-broadway/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M
```

Lemonade

How to use john-broadway/Qwen3-32B-RYS-20-28-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull john-broadway/Qwen3-32B-RYS-20-28-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3-32B-RYS-20-28-GGUF-Q4_K_M

List all available models

lemonade list

Qwen3-32B-RYS-20-28

RYS-enhanced Qwen3-32B with layers 20-28 duplicated. 64 layers expanded to 72. Zero training, zero weight changes.

Math +4.5%, EQ +0.04, Reasoning +18%. All three metrics positive.

Results

Metric	Baseline	RYS (20,28)	Delta
Math	0.7525	0.7974	+4.5%
EQ	94.88	94.92	+0.04
Reasoning	70.59%	88.24%	+18%

The Big Boy. Baseline logic score was 0% — RYS fixes it. The (20,28) config is the only one that improves all three metrics simultaneously.

Usage

llama-server -m Qwen3-32B-RYS-20-28-Q4_K_M.gguf -ngl 99

Full sweep data

63 configurations tested. Sweep results published with the model files.

Part of the RYS Sovereign Collection v1 — the headline crossover.

Where this sits in the Sovereign Collection

v1 — Qwen2.5 cross-scale + Qwen3-32B headline crossover. Four sizes from 0.5B to 32B; RYS works at every scale, with the lift size and dimension shifting by baseline:

0.5B → EQ specialist (Qwen2.5-0.5B-RYS-3-7-GGUF)
1.5B → balanced daily driver (Qwen2.5-1.5B-RYS-4-7-GGUF)
7B → math specialist via (8,12) (Qwen2.5-7B-RYS-8-12-GGUF + -AWQ)
32B (this card) — the headline "Big Boy"

v1 attribution: John Broadway, with collaboration from Claude (Opus 4.6 in April 2026 build; Opus 4.7 in May 2026 v1 republication aligning to the original Qwen2.5 cross-scale intent). Original RYS method by David Ng on Qwen2-72B; sweep toolkit by alainnothere.

v2 cross-architecture context (2026-05-13)

This model's place in the v2 curve: baseline reasoning 70.59%, peak RYS Δ +17.65%. The (20,28) 8-layer block sits at 30–50% depth, the family-invariant band for deeper models. 38 boosting configurations make this the most-validated single-model dataset in v2.

Across the 21 model variants (10 architecture families) surveyed in john-broadway/rys-sovereign-collection-v2:

Pearson r(baseline reasoning, peak RYS lift) = −0.726. Weak baselines lift more, in their weakest dimension.
Three RYS-recoverable suppression mechanisms identified: under-training scale, MoE routing inefficiency, specialization training trade-off.
One published negative result (SmolLM2-1.7B). RYS is not universal.

v2 attribution: John Broadway, with cross-architecture analysis by Claude (Opus 4.7). Original RYS method by David Ng; circuit-finder toolkit by alainnothere.

Downloads last month: 199

GGUF

Model size

37B params

Architecture

qwen3

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for john-broadway/Qwen3-32B-RYS-20-28-GGUF

Base model

Qwen/Qwen3-32B

Quantized

(152)

this model

Collection including john-broadway/Qwen3-32B-RYS-20-28-GGUF

RYS Sovereign Collection v1

Collection

Qwen2.5 cross-scale (0.5/1.5/7B) + Qwen3-32B headline. 5 RYS-applied model repos. v2 collection: Qwen3 cohort + 21-variant corpus. • 5 items • Updated 7 days ago