Instructions to use Rangle2/Wit-Coder-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Rangle2/Wit-Coder-4B 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 Rangle2/Wit-Coder-4B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Rangle2/Wit-Coder-4B:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Rangle2/Wit-Coder-4B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Rangle2/Wit-Coder-4B: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 Rangle2/Wit-Coder-4B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf Rangle2/Wit-Coder-4B:Q4_K_M

Use Docker

docker model run hf.co/Rangle2/Wit-Coder-4B:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use Rangle2/Wit-Coder-4B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Rangle2/Wit-Coder-4B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Rangle2/Wit-Coder-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Rangle2/Wit-Coder-4B:Q4_K_M

Ollama
How to use Rangle2/Wit-Coder-4B with Ollama:
```
ollama run hf.co/Rangle2/Wit-Coder-4B:Q4_K_M
```
Unsloth Desktop

How to use Rangle2/Wit-Coder-4B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Rangle2/Wit-Coder-4B: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": "Rangle2/Wit-Coder-4B:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Rangle2/Wit-Coder-4B with Docker Model Runner:
```
docker model run hf.co/Rangle2/Wit-Coder-4B:Q4_K_M
```

Lemonade

How to use Rangle2/Wit-Coder-4B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Rangle2/Wit-Coder-4B:Q4_K_M

Run and chat with the model

lemonade run user.Wit-Coder-4B-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Rangle2/Wit-Coder-4B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Rangle2/Wit-Coder-4B: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 Rangle2/Wit-Coder-4B:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Rangle2/Wit-Coder-4B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Rangle2/Wit-Coder-4B: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 "Rangle2/Wit-Coder-4B: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"

Wit-Coder-4B

File size: 3,514 Bytes

df5b3c0

---
license: apache-2.0
base_model: Qwen/Qwen3.5-4B
language:
  - en
pipeline_tag: text-generation
tags:
  - code
  - agent
  - dpo
  - gguf
---

# Wit-Coder-4B

An English coding-agent fine-tune of [Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B), focused on completing solvable tasks,
breaking repetitive actions, and explaining blockers. This release candidate uses **DPO-2**.

## Results

Project measurements of the base model and DPO-2:

| Benchmark | Qwen3.5-4B | Wit-Coder-4B |
|---|---:|---:|
| MMLU, 5-shot | 69.67% | 70.16% |
| GSM8K, 8-shot CoT, strict | 84.08% | 82.64% |
| HumanEval plus tests, modified bounded protocol | 81.10% (133/164) | 87.20% (143/164) |
| MATH-500, first 100 problems | 65/100 | 82/100 |

GSM8K decreased. MBPP results are unavailable. **HumanEval scores are not official EvalPlus results:** two tasks were chunked,
resetting candidate state; no HumanEval inputs were omitted. Canonical HumanEval/32 also failed validation in this environment.
Benchmark scores do not measure the quantized GGUF files. See [evaluation details](EVALUATION.md) for settings and limitations.

### Coding-agent panel

42 tasks × 3 seeds, including 90 solvable and 36 blocked episodes:

| Outcome | Base | Wit |
|---|---:|---:|
| Task success, including successful escalation | 40/126 | 71/126 |
| Blocked tasks successfully escalated | 0/36 | 11/36 |
| Protected-file integrity failures | 28/126 | 25/126 |

This small panel informed later experiments and model selection; it is a development comparison, not an independent final test.
False completion remains a problem: 15/36 blocked episodes ended with a non-escalating submit. Use external tests and tool permissions.
Only English text-based coding-agent use was evaluated; no broad chat, vision or long-context claim is made.

## Training

Approximately 138M SFT tokens, then two sandbox-graded DPO rounds of 877 and 404 turn-level pairs.
Round 2 used self-check-guided candidate sampling. GRPO checkpoints were not selected for release because the conservative run showed no clear improvement over DPO-2. A more aggressive run increased looping and reduced escalation on the monitoring set, so it was stopped.

## Usage

```bash
vllm serve /path/to/Wit-Coder-4B --max-model-len 32768 --reasoning-parser qwen3
```

The training harness used one `bash` tool with Qwen-style XML calls and task-specific submit/escalation markers;
see [details](EVALUATION.md). Adapt tool handling to your application.

### GGUF

| Format | File | Size |
|---|---|---:|
| BF16 | [Download](https://huggingface.co/Rangle2/Wit-Coder-4B/resolve/main/gguf/Wit-Coder-4B-bf16.gguf) | 8.67 GB |
| Q8_0 | [Download](https://huggingface.co/Rangle2/Wit-Coder-4B/resolve/main/gguf/Wit-Coder-4B-Q8_0.gguf) | 4.61 GB |
| Q4_K_M | [Download](https://huggingface.co/Rangle2/Wit-Coder-4B/resolve/main/gguf/Wit-Coder-4B-Q4_K_M.gguf) | 2.78 GB |

Available formats: BF16, Q8_0 and Q4_K_M.
Checksums/settings: `gguf/manifest.json`. Structure and hashes were checked; GGUF inference and quantized quality were not evaluated.
No image projector is supplied.

## Support

I research alignment, internal representations, and behavioral fine-tuning of small language models.
I am preparing for military service; optional [Buy Me a Coffee support](https://buymeacoffee.com/rangle2)
helps fund future independent projects when I return. Wit remains free under Apache-2.0.

## License

[Apache-2.0](https://huggingface.co/Qwen/Qwen3.5-4B/blob/main/LICENSE), following the base model.