Instructions to use Xx-Vexento-xX/openhome-nova with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Xx-Vexento-xX/openhome-nova with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-0.5B-Instruct")
model = PeftModel.from_pretrained(base_model, "Xx-Vexento-xX/openhome-nova")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use Xx-Vexento-xX/openhome-nova 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 Xx-Vexento-xX/openhome-nova
# Run inference directly in the terminal:
llama cli -hf Xx-Vexento-xX/openhome-nova

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Xx-Vexento-xX/openhome-nova
# Run inference directly in the terminal:
llama cli -hf Xx-Vexento-xX/openhome-nova

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 Xx-Vexento-xX/openhome-nova
# Run inference directly in the terminal:
./llama-cli -hf Xx-Vexento-xX/openhome-nova

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 Xx-Vexento-xX/openhome-nova
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Xx-Vexento-xX/openhome-nova

Use Docker

docker model run hf.co/Xx-Vexento-xX/openhome-nova

LM Studio
Jan

vLLM

How to use Xx-Vexento-xX/openhome-nova with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Xx-Vexento-xX/openhome-nova

Ollama
How to use Xx-Vexento-xX/openhome-nova with Ollama:
```
ollama run hf.co/Xx-Vexento-xX/openhome-nova
```

Unsloth Studio

How to use Xx-Vexento-xX/openhome-nova 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 Xx-Vexento-xX/openhome-nova 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 Xx-Vexento-xX/openhome-nova to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Xx-Vexento-xX/openhome-nova to start chatting

How to use Xx-Vexento-xX/openhome-nova with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Xx-Vexento-xX/openhome-nova

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": "Xx-Vexento-xX/openhome-nova"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Xx-Vexento-xX/openhome-nova with Docker Model Runner:
```
docker model run hf.co/Xx-Vexento-xX/openhome-nova
```

Lemonade

How to use Xx-Vexento-xX/openhome-nova with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Xx-Vexento-xX/openhome-nova

Run and chat with the model

lemonade run user.openhome-nova-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use Xx-Vexento-xX/openhome-nova with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Xx-Vexento-xX/openhome-nova

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 Xx-Vexento-xX/openhome-nova

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Xx-Vexento-xX/openhome-nova with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Xx-Vexento-xX/openhome-nova

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 "Xx-Vexento-xX/openhome-nova" \
  --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"

openhome-nova

LoRA fine-tune of Qwen/Qwen2.5-0.5B-Instruct for smart home event reasoning and action dispatch. Built for OpenHome — a fully local, open-source smart home security system running on a Raspberry Pi with no cloud dependency.

What it does

Given a sensor event (door opened, motion detected, smoke alarm, etc.) and the current state of all smart home devices, NOVA returns a structured JSON action list:

[
  {"action": "notify", "message": "Smoke detected in kitchen — evacuate!", "severity": "critical"},
  {"action": "all_lights", "mode": "alert"}
]

It handles:

Sensor events — panic, smoke/CO, flood, motion, door, doorbell, RFID, climate, air quality
Natural language commands — "goodnight", "movie mode", "I'm cold", "lock everything down"
Routines — away mode, lockdown, morning, party, arrive-home sequences
Constraints — refuses to write read-only sensors, won't invent device IDs

Intended use

Deployed in OpenHome via Ollama on a Raspberry Pi. The hub loads NOVA as the reasoning layer between raw sensor events and actuator commands.

Load it:

# After cloning the OpenHome repo and downloading this adapter to openhome-model/lora/
ollama create openhome-nova -f pi/Modelfile

Or load directly with PEFT:

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-0.5B-Instruct")
model = PeftModel.from_pretrained(base, "Xx-Vexento-xX/openhome-nova")
tok = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-0.5B-Instruct")

Training details


Base model	Qwen/Qwen2.5-0.5B-Instruct
Method	LoRA (PEFT)
LoRA rank	8
LoRA alpha	16
Target modules	q_proj, v_proj
Training examples	1718
Validation examples	191
Epochs	2
Learning rate	2e-4
Batch size	1 (gradient accumulation 4)
Max sequence length	1024 (truncation_side=left)
Hardware	CPU only — AMD Ryzen 7 5800X3D, 34 GB RAM
Training time	~9 hours
Final eval loss	0.044

Training was done entirely on CPU using transformers + peft + trl (no CUDA, no Unsloth). The truncation_side="left" setting was critical to preserve the assistant's response tokens when sequences exceeded max length.

Evaluation

Tested against 10 OpenHome scenarios immediately after training:

Scenario	Result
Panic button pressed	PASS
Smoke alarm triggered	PASS
Lights off command	PASS
Goodnight routine	PASS
Thermostat adjustment	PASS
Flood sensor wet	PASS
No-op (no action needed)	PASS
Door open at night	PASS
Movie mode	PASS
I'm home routine	PASS

10/10 valid JSON, 9/10 action-correct (90%). The one partial-pass was smoke/CO outputting "alert" instead of "notify" as the action type — handled in the OpenHome receiver with an alias rather than requiring retraining.

Training data

The dataset covers 18 categories. Key examples:

All 13 device types with correct sensor → action mappings
Time-of-day logic (door open at night = alert, daytime = log)
Light commands: 200+ variations (color, brightness, mode, all-lights)
Thermostat commands: 200+ (setpoint, mode, hold, schedule)
Household routines: goodnight, morning, movie, party, lockdown, away, arrive
100 constraint examples — teaching what the model must NOT do (can't unlock RFID, can't clear smoke alarm, can't write read-only sensors)
50 no-op examples — so it returns [] instead of inventing actions for benign events

Regenerate or expand: python3 dataset/generate_training_data.py in the OpenHome repo.

Action schema

NOVA outputs arrays of these action types:

{"action": "control_device", "device_id": "plug_kitchen", "state": "on|off|toggle"}
{"action": "set_thermostat", "device_id": "thermostat_main", "setpoint": 72, "mode": "heat|cool|auto|off"}
{"action": "set_light_mode", "device_id": "rgb_living", "mode": "alert|rainbow|pulse|fire|solid", "color": "#FF0000"}
{"action": "notify", "message": "...", "severity": "low|medium|high|critical"}
{"action": "all_lights", "mode": "alert|off|on"}
{"action": "log", "message": "..."}

Limitations

Trained for the OpenHome device schema. Device type names and field names must match the OpenHome standard for best results.
At 0.5B parameters, complex multi-step reasoning can occasionally miss edge cases — the RAG memory layer in OpenHome compensates with household context.
Does not handle arbitrary home automation platforms (Home Assistant, SmartThings, etc.) out of the box.

Repository

Full system source, ESP32 firmware, Pi hub, dashboard, and training pipeline: github.com/DaEpickid540/OpenHome

License

MIT

Downloads last month: 3

Safetensors

Model size

0.5B params

Tensor type

F32

BF16

Model tree for Xx-Vexento-xX/openhome-nova

Base model

Qwen/Qwen2.5-0.5B

Finetuned

Qwen/Qwen2.5-0.5B-Instruct

Adapter

(738)

this model