Instructions to use josephmayo/von3b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use josephmayo/von3b 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 josephmayo/von3b:Q8_0
# Run inference directly in the terminal:
llama cli -hf josephmayo/von3b:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf josephmayo/von3b:Q8_0
# Run inference directly in the terminal:
llama cli -hf josephmayo/von3b:Q8_0

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 josephmayo/von3b:Q8_0
# Run inference directly in the terminal:
./build/bin/llama cli -hf josephmayo/von3b:Q8_0

Use Docker

docker model run hf.co/josephmayo/von3b:Q8_0

Llama new
LM Studio
Jan
Ollama
How to use josephmayo/von3b with Ollama:
```
ollama run hf.co/josephmayo/von3b:Q8_0
```
Unsloth Desktop

How to use josephmayo/von3b with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf josephmayo/von3b:Q8_0

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": "josephmayo/von3b:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use josephmayo/von3b with Docker Model Runner:
```
docker model run hf.co/josephmayo/von3b:Q8_0
```

Lemonade

How to use josephmayo/von3b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull josephmayo/von3b:Q8_0

Run and chat with the model

lemonade run user.von3b-Q8_0

List all available models

lemonade list

Hermes Agent

How to use josephmayo/von3b with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf josephmayo/von3b:Q8_0

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 josephmayo/von3b:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use josephmayo/von3b with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf josephmayo/von3b:Q8_0

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 "josephmayo/von3b:Q8_0" \
  --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"

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

VON-3B

This is a hackathon submission repo and my 3B model. designed and ran the full path: SFT, RL, group-conditioned adaptive LoPD (rl & distillation method from my research), LoRA, weight edits, and the laptop GGUF pack.

VON-3B is one 3B model that covers two jobs: an offline coding assistant and an autonomous agent. after one public download it runs locally in llama.cpp or any inference engine on a standard 8 GB machine. no API key / network at inference.

we built it to:

write and repair code
keep reasoning short (base reasoned too long and took lots of time)
emit a real one-line tool call (<tool_call>{...}</tool_call>) so it can act as an agent, not only a chatbot

other details:

model card and weights: https://huggingface.co/josephmayo/von3b
laptop artifact: von3b-Q8_0.gguf (llama.cpp, GGUF Q8_0, 3,285,475,488 bytes)
runtime: llama.cpp only
target machine: 4 vCPU, 8 GB RAM, integrated GPU, Ubuntu 22.04
writeup: REPORT.md

we started from WeiboAI/VibeThinker-3B. On a matched EvalPlus 0.3.1 HumanEval check (same 164 tasks, greedy max_new=8192), the model beats that snapshot:

Arm	HumanEval pass@1	HumanEval+ pass@1
VON-3B	0.921 (151 / 164)	0.884 (145 / 164)
VibeThinker-3B base	0.866 (142 / 164)	0.817 (134 / 164)

we report that comparison because we were compute-constrained - thats why we didnt run more evals. tool probe, 32 tasks, greedy max_new=256, same snapshot: the model emits a valid one-line <tool_call> with short think on 32 / 32. The base emits 0 / 32.

download:

bash download_model.sh
llama-cli -m model/von3b-Q8_0.gguf -c 65536 -ctk q4_0 -ctv q4_0 -ngl 0

ctx is 65,536 with Q4_0 K/V cache so the 8 GB profile can hold long coding sessions.

required package files (official template):

metadata.json
download_model.sh
REPORT.md
model/ (GGUF downloaded by the script, not committed)

Downloads last month: 119

Safetensors

Model size

3B params

Tensor type

F16

Inference Providers NEW

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