Instructions to use rrchat/Cerberus-E4B-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 rrchat/Cerberus-E4B-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 rrchat/Cerberus-E4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf rrchat/Cerberus-E4B-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 rrchat/Cerberus-E4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf rrchat/Cerberus-E4B-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf rrchat/Cerberus-E4B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf rrchat/Cerberus-E4B-GGUF:Q4_K_M

Use Docker

docker model run hf.co/rrchat/Cerberus-E4B-GGUF:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use rrchat/Cerberus-E4B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "rrchat/Cerberus-E4B-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": "rrchat/Cerberus-E4B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/rrchat/Cerberus-E4B-GGUF:Q4_K_M

Ollama
How to use rrchat/Cerberus-E4B-GGUF with Ollama:
```
ollama run hf.co/rrchat/Cerberus-E4B-GGUF:Q4_K_M
```
Unsloth Desktop

How to use rrchat/Cerberus-E4B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf rrchat/Cerberus-E4B-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": "rrchat/Cerberus-E4B-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use rrchat/Cerberus-E4B-GGUF with Docker Model Runner:
```
docker model run hf.co/rrchat/Cerberus-E4B-GGUF:Q4_K_M
```

Lemonade

How to use rrchat/Cerberus-E4B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull rrchat/Cerberus-E4B-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Cerberus-E4B-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use rrchat/Cerberus-E4B-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 rrchat/Cerberus-E4B-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 rrchat/Cerberus-E4B-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use rrchat/Cerberus-E4B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf rrchat/Cerberus-E4B-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 "rrchat/Cerberus-E4B-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"

Cerberus-E4B-GGUF: Multimodal & Multi-Task Reasoning

This repository provides official quantized GGUF models for Cerberus-E4B, an open-weights multi-task reasoning, agentic and multimodal model fine-tuned on top of Google's gemma-4-E4B-it architecture (~7.5B total parameters with frozen vision and audio towers, 4B active language backbone).

Looking for the original adapter weights?
👉 Cerberus-E4B

Available Files & Quantizations

File	Size	VRAM Req.	Recommended Use Case
`cerberus-q4_k_m.gguf`	4.97 GB	6–8 GB	Recommended (Best balance of speed, size, and perplexity)
`cerberus-q5_k_m.gguf`	5.37 GB	8 GB	Higher precision math and code generation
`cerberus-q8_0.gguf`	7.48 GB	12 GB+	Near-lossless 8-bit quantization
`cerberus-bf16.gguf`	14.02 GB	16 GB+	Uncompressed 16-bit floating point master
`mmproj-cerberus-bf16.gguf`	991.5 MB	~1 GB	Multimodal Projector (Vision ViT 280 tokens + Audio Conformer 16kHz)
`mmproj-cerberus-f16.gguf`	990.4 MB	~1 GB	Multimodal Projector (Float16 variant for specific backends)

Benchmark Highlights (vs. Base Model)

Benchmark	Base Model (`Gemma-4-E4B-it`)	Cerberus-E4B	Delta (vs. Base)	Generation Mode
MATH-500	43.3%	60.0%	+16.7%	Thinking Mode (`<thought>`)
MATH-500	35.0%	48.0%	+13.0%	Instant Mode (Direct calculation)
MMLU-Redux	62.0%	63.0%	+1.0%	Multi-discipline academic knowledge
IFEval	90.8%	90.5%	Parity (~90%)	Complex system constraint following

How to Run

1. In llama.cpp CLI (with Vision or Audio)

# Vision (Image understanding):
llama-mtmd-cli \
  -m cerberus-q4_k_m.gguf \
  --mmproj mmproj-cerberus-bf16.gguf \
  --image sample.jpg \
  -c 8192 \
  -p "Describe the details in this image."

# Audio (Speech recognition / question answering):
llama-mtmd-cli \
  -m cerberus-q4_k_m.gguf \
  --mmproj mmproj-cerberus-bf16.gguf \
  --audio speech.wav \
  -c 8192 \
  -p "Transcribe the audio and answer the question."

2. In llama.cpp Web Server (Browser Chat UI)

llama-server \
  -m cerberus-q4_k_m.gguf \
  --mmproj mmproj-cerberus-bf16.gguf \
  -ngl 99 \
  -c 8192 \
  --port 8888

Open http://localhost:8888 in your browser. You can drag and drop images and audio files directly into the chat interface!

3. In Ollama

Because this repository contains an official Modelfile, you can run it directly:

# Direct run from Hugging Face:
ollama run hf.co/rrlabs-ai/Cerberus-E4B-GGUF

Or build a local model using the provided Modelfile:

ollama create cerberus -f Modelfile
ollama run cerberus

4. In LM Studio

Place both cerberus-q4_k_m.gguf and mmproj-cerberus-bf16.gguf in your LM Studio models folder.
Select cerberus-q4_k_m.gguf.
In the sidebar under Vision Adapter, choose mmproj-cerberus-bf16.gguf.

5. In rrchat (Recommended Mobile & Web UI)

Cerberus-E4B-GGUF is the ideal model to power rrchat:

Use your local PC's llama.cpp from your phone. Clean mobile-ready AI chat interface with vision, web search, and privacy.

🌐 Project Repository: github.com/leftik1/rrchat
🚀 How to run:
1. Start llama-server on your PC with Cerberus GGUF and mmproj:
```
llama-server -m cerberus-q4_k_m.gguf --mmproj mmproj-cerberus-bf16.gguf -ngl 99 -c 8192 --port 8888 --host 0.0.0.0
```
2. Open rrchat on your smartphone or browser. You can snap and upload photos from your phone camera for instant multimodal answers, search the web, and chat with 100% privacy!

Recommended Sampling Settings

Thinking Mode (<|channel>thought):
- Use temperature 0.8 (or 1.0) with top_p: 0.95, top_k: 64.
- Enable the DRY sampler (--dry-multiplier 0.8 --dry-base 1.75 --dry-allowed-length 2) in llama.cpp to ensure long thought traces do not loop.
Instant Mode / Code / Tool Calling:
- Use pure greedy decoding (temperature: 0.0) for maximum determinism and structured JSON fidelity.

Dataset Attribution & Acknowledgements

NVIDIA Nemotron: Derived in part from nvidia/Nemotron-Post-Training-Dataset-v2 and nvidia/Nemotron-SFT-Agentic-v2, licensed under the NVIDIA Open Synthetic Data License.
Hugging Face Open-R1: Mathematical logic traces from open-r1/OpenR1-Math-220k (Apache-2.0).
Hugging Face SmolTalk: Conversational data from HuggingFaceTB/smoltalk2 (Apache-2.0).
Vikhr Models: Russian-language reasoning traces from Vikhrmodels/GrandMaster-PRO-MAX (Apache-2.0).
Base Architecture: Google DeepMind's google/gemma-4-E4B-it (Gemma Terms of Use).

License

Released under the Apache-2.0 License. Use of the underlying model remains subject to the Gemma Terms of Use.

Downloads last month: 896

GGUF

Model size

8B params

Architecture

gemma4

Hardware compatibility

4-bit

5-bit

8-bit

16-bit

Model tree for rrchat/Cerberus-E4B-GGUF

Base model

google/gemma-4-E4B

Finetuned

google/gemma-4-E4B-it

Quantized

(426)

this model

Space using rrchat/Cerberus-E4B-GGUF 1

Collection including rrchat/Cerberus-E4B-GGUF

RR-Gen1

Collection

Our first generation of open-weight models • 6 items • Updated 3 days ago • 1