Instructions to use deucebucket/Qwen3-32B-Cerebellum-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use deucebucket/Qwen3-32B-Cerebellum-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="deucebucket/Qwen3-32B-Cerebellum-GGUF",
	filename="Qwen3-32B-Cerebellum-v1.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 deucebucket/Qwen3-32B-Cerebellum-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf deucebucket/Qwen3-32B-Cerebellum-GGUF
# Run inference directly in the terminal:
llama-cli -hf deucebucket/Qwen3-32B-Cerebellum-GGUF

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf deucebucket/Qwen3-32B-Cerebellum-GGUF
# Run inference directly in the terminal:
llama-cli -hf deucebucket/Qwen3-32B-Cerebellum-GGUF

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 deucebucket/Qwen3-32B-Cerebellum-GGUF
# Run inference directly in the terminal:
./llama-cli -hf deucebucket/Qwen3-32B-Cerebellum-GGUF

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 deucebucket/Qwen3-32B-Cerebellum-GGUF
# Run inference directly in the terminal:
./build/bin/llama-cli -hf deucebucket/Qwen3-32B-Cerebellum-GGUF

Use Docker

docker model run hf.co/deucebucket/Qwen3-32B-Cerebellum-GGUF

LM Studio
Jan
Ollama
How to use deucebucket/Qwen3-32B-Cerebellum-GGUF with Ollama:
```
ollama run hf.co/deucebucket/Qwen3-32B-Cerebellum-GGUF
```

Unsloth Studio new

How to use deucebucket/Qwen3-32B-Cerebellum-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 deucebucket/Qwen3-32B-Cerebellum-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 deucebucket/Qwen3-32B-Cerebellum-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for deucebucket/Qwen3-32B-Cerebellum-GGUF to start chatting

Pi new

How to use deucebucket/Qwen3-32B-Cerebellum-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf deucebucket/Qwen3-32B-Cerebellum-GGUF

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": "deucebucket/Qwen3-32B-Cerebellum-GGUF"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use deucebucket/Qwen3-32B-Cerebellum-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 deucebucket/Qwen3-32B-Cerebellum-GGUF

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 deucebucket/Qwen3-32B-Cerebellum-GGUF

Run Hermes

hermes

Docker Model Runner
How to use deucebucket/Qwen3-32B-Cerebellum-GGUF with Docker Model Runner:
```
docker model run hf.co/deucebucket/Qwen3-32B-Cerebellum-GGUF
```

Lemonade

How to use deucebucket/Qwen3-32B-Cerebellum-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull deucebucket/Qwen3-32B-Cerebellum-GGUF

Run and chat with the model

lemonade run user.Qwen3-32B-Cerebellum-GGUF-{{QUANT_TAG}}

List all available models

lemonade list

Qwen3-32B — Cerebellum GGUF

Ablation-guided mixed-precision quantization of Qwen/Qwen3-32B. 32B parameters, dense architecture with GQA (64 heads, 8 KV heads), 64 layers.

What is Cerebellum?

Instead of uniform quantization, we measure which weight groups survive aggressive compression and which don't. Groups that tolerate Q2_K get demoted; groups that don't stay at Q3_K_M or higher. The result: smaller files with less quality loss than uniform quants of the same size.

Files

File	Size	Description
`Qwen3-32B-Cerebellum-v2.gguf`	15 GB	Optimal mix — 3 groups demoted (attn_k, attn_q, attn_output), 4 kept at Q3_K_M
`Qwen3-32B-Cerebellum-v1.gguf`	14 GB	Aggressive — 5 groups demoted (all attn + ffn_gate)

Benchmarks

Evaluated using our standardized benchmark suite with temperature=0, no thinking mode.

Cerebellum v2 (15 GB) — Recommended

Benchmark	Score	Questions
ARC-Challenge	92.8%	1,172
HellaSwag	87.4%	10,042
MMLU	75.5%	11,643
HumanEval	45.1%	164

Note: HumanEval score reflects non-thinking mode. Qwen3 models perform significantly better on code with thinking enabled (/think or enable_thinking: true).

Size vs Quality

Model	Size	BPW	PPL (wiki)
Q3_K_M (baseline)	16 GB	3.94	8.3288
Cerebellum v2	15 GB	3.67	8.3435
Cerebellum v1	14 GB	3.44	8.7273

v2 saves 1 GB (6%) over Q3_K_M with only +0.18% perplexity increase — essentially lossless. The attn_k group actually improved perplexity when demoted.

Methodology

Group ablation: Demote each of 7 weight groups to Q2_K individually. Measure PPL impact.
Identify cheap groups: Three groups (attn_k, attn_q, attn_output) showed negligible or negative PPL impact when demoted.
Build optimal mix: v2 demotes the 3 cheapest groups; v1 additionally demotes attn_v and ffn_gate.

Ablation Results

Group	PPL when demoted	Delta vs baseline
attn_k	8.3019	-0.0269 (improved!)
attn_q	8.3394	+0.0106
attn_output	8.3406	+0.0118
attn_v	8.3700	+0.0412
ffn_gate	8.6159	+0.2871
ffn_up	8.7267	+0.3979
ffn_down	8.8200	+0.4912

v2 Override Map

Demoted (Q2_K): attn_k, attn_q, attn_output (all 64 layers)
Sacred (kept at Q3_K_M): attn_v, ffn_gate, ffn_up, ffn_down

Usage

Works with any llama.cpp-compatible tool:

# llama.cpp
./llama-server --model Qwen3-32B-Cerebellum-v2.gguf -ngl 99 --ctx-size 4096

# Ollama (create Modelfile pointing to the GGUF)
# LM Studio (drag and drop)
# koboldcpp, text-generation-webui, etc.

Hardware Requirements

v2 (15 GB): Fits in 24 GB VRAM with generous context. RTX 3090, RTX 4090, etc.
v1 (14 GB): Fits in 16 GB VRAM with limited context. RTX 4060 Ti 16GB, etc.

Credits

Quantized with Cerebellum — ablation-guided mixed-precision quantization by deucebucket.
Base model by Qwen.

Downloads last month: 404

GGUF

Model size

33B params

Architecture

qwen3

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Inference Providers NEW

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

Model tree for deucebucket/Qwen3-32B-Cerebellum-GGUF

Base model

Qwen/Qwen3-32B

Quantized

(152)

this model