Instructions to use NewSonnet/omarchy-nano-2b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use NewSonnet/omarchy-nano-2b with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="NewSonnet/omarchy-nano-2b")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("NewSonnet/omarchy-nano-2b", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use NewSonnet/omarchy-nano-2b 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 NewSonnet/omarchy-nano-2b:F16
# Run inference directly in the terminal:
llama cli -hf NewSonnet/omarchy-nano-2b:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf NewSonnet/omarchy-nano-2b:F16
# Run inference directly in the terminal:
llama cli -hf NewSonnet/omarchy-nano-2b:F16

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 NewSonnet/omarchy-nano-2b:F16
# Run inference directly in the terminal:
./llama-cli -hf NewSonnet/omarchy-nano-2b:F16

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 NewSonnet/omarchy-nano-2b:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf NewSonnet/omarchy-nano-2b:F16

Use Docker

docker model run hf.co/NewSonnet/omarchy-nano-2b:F16

LM Studio
Jan

vLLM

How to use NewSonnet/omarchy-nano-2b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "NewSonnet/omarchy-nano-2b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NewSonnet/omarchy-nano-2b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/NewSonnet/omarchy-nano-2b:F16

SGLang

How to use NewSonnet/omarchy-nano-2b with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "NewSonnet/omarchy-nano-2b" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NewSonnet/omarchy-nano-2b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "NewSonnet/omarchy-nano-2b" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NewSonnet/omarchy-nano-2b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use NewSonnet/omarchy-nano-2b with Ollama:
```
ollama run hf.co/NewSonnet/omarchy-nano-2b:F16
```
Unsloth Desktop

How to use NewSonnet/omarchy-nano-2b with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf NewSonnet/omarchy-nano-2b:F16

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": "NewSonnet/omarchy-nano-2b:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use NewSonnet/omarchy-nano-2b with Docker Model Runner:
```
docker model run hf.co/NewSonnet/omarchy-nano-2b:F16
```

Lemonade

How to use NewSonnet/omarchy-nano-2b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull NewSonnet/omarchy-nano-2b:F16

Run and chat with the model

lemonade run user.omarchy-nano-2b-F16

List all available models

lemonade list

Hermes Agent

How to use NewSonnet/omarchy-nano-2b with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf NewSonnet/omarchy-nano-2b:F16

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 NewSonnet/omarchy-nano-2b:F16

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use NewSonnet/omarchy-nano-2b with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf NewSonnet/omarchy-nano-2b:F16

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 "NewSonnet/omarchy-nano-2b:F16" \
  --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"

Omarchy Nano 2B

Omarchy Nano 2B is an experimental, text-only fine-tune of Qwen3.5-2B-Base for questions about Omarchy. It focuses on Omarchy configuration and related command-line workflows. It is an independent community model, not an official Omarchy release.

Use it as a local reference assistant. Treat its output as unverified guidance and check commands against current Omarchy documentation before running them, especially before making system or security-sensitive changes.

Choose an artifact

Use case	File	Approximate size	Notes
Recommended local text inference	`gguf-q4_k_m/omarchy-nano.Q4_K_M.gguf`	1.31 GB	Quantized GGUF for `llama.cpp`-compatible runners.
Optional multimodal runner support	`gguf-q4_k_m/omarchy-nano.F16-mmproj.gguf`	668 MB	Download only when your runner requires a projector for image input.
PEFT/QLoRA research	`adapter/`	43.7 MB adapter weights plus tokenizer files	Requires the original Qwen base model; it is not standalone.

For a low-memory PC, download only the Q4_K_M GGUF. The fine-tuning data was text-only. The presence of a projector in the repository does not mean that this fine-tune learned image understanding.

Quick start with llama.cpp

wget -O omarchy-nano.Q4_K_M.gguf \
  https://huggingface.co/NewSonnet/omarchy-nano-2b/resolve/main/gguf-q4_k_m/omarchy-nano.Q4_K_M.gguf

llama-cli -m omarchy-nano.Q4_K_M.gguf -cnv

Use a recent llama.cpp build with GGUF and Qwen3.5 support. If your runner does not automatically select the embedded chat template, enable its Jinja or chat-template option according to that runner's documentation.

What was trained

The training corpus contains 296 examples from 51 non-empty Markdown documents in the Omarchy manual.

Training split: 259 examples
Evaluation split: 37 examples
Split method: document-level
Modality: text-only
Image references replaced with: [image omitted]
Omarchy source commit: d3d23fdd

The reproducibility repository contains the reproducibility record for this run. The raw training corpus is not bundled in this model repository.

Training configuration

Method: 4-bit QLoRA
LoRA rank: 16
Trainable parameters: 10,911,744 (0.49%)
Epochs: 2
Maximum sequence length: 512
Effective batch size: 8
Learning rate: 2e-4
Optimizer: 8-bit AdamW
Hardware: Google Colab Tesla T4

Final optimization metrics:

Metric	Value
Training loss	1.6297
Held-out evaluation loss	1.6660

These are language-model optimization metrics, not task-accuracy or command-safety results. Because the held-out set comes from the same manual as the training data, it is not an independent measure of real-world Omarchy support quality.

Adapter use

The adapter/ directory contains a PEFT LoRA adapter plus the tokenizer and processor files saved with the run. Apply it to Qwen/Qwen3.5-2B-Base; it is not a complete model and cannot be loaded by itself.

The adapter directory includes a machine-readable training_summary.json and a dedicated adapter guide. Use a current Transformers and PEFT installation compatible with Qwen3.5.

Limitations and out-of-scope use

This is a small, narrowly focused model. Omarchy changes over time, so its coverage may become outdated.
It can produce incorrect commands or configuration details. Inspect every generated command before execution.
No independent benchmark of answer quality, command correctness, or safety was run for this release.
Do not give it unrestricted access to the host system or external tools.
Training used text only; the model does not interpret visual input.

Reproducibility

The companion repository provides a cleaned, credential-free Colab notebook and the files needed to reproduce preprocessing and training:

The notebook's default path is local to the Colab runtime: it does not require Google Drive permission, Hugging Face authentication, or publishing. Drive storage and Hub publishing are optional settings for users who need durable outputs or a private/public copy of their own run.

Licensing and attribution

The Qwen3.5-2B-Base model and the model artifacts derived from it are provided under the Apache License 2.0. The Omarchy source material used to create the training examples is released under the MIT License. Users must comply with both licenses and retain the relevant attribution. The license metadata above describes the Qwen-derived model artifact; it does not change the license of the Omarchy source material.

Model tree for NewSonnet/omarchy-nano-2b

Base model

Qwen/Qwen3.5-2B-Base

Finetuned

(120)

this model

NewSonnet
/

omarchy-nano-2b

Omarchy Nano 2B

Choose an artifact

Quick start with llama.cpp

What was trained

Training configuration

Adapter use

Limitations and out-of-scope use

Reproducibility

Licensing and attribution

Links

Model tree for NewSonnet/omarchy-nano-2b