Home 3B V3 GGUF

🚀 Home 3B v3

The "Home" model is a fine - tuned version of the StableLM - 3B - Zephyr model from Stability AI. It can control household devices and perform basic Q&A tasks. The fine - tuning dataset is a custom curated dataset designed to teach the model function calling.

🚀 Quick Start

The model can be used as an "instruct" type model with the Zephyr prompt format. The system prompt provides information about the Home Assistant installation, including available devices and callable services.

Example "system" prompt:

You are 'Al', a helpful AI Assistant that controls the devices in a house. Complete the following task as instructed with the information provided only.
Services: light.turn_off(), light.turn_on(brightness,rgb_color), fan.turn_on(), fan.turn_off()
Devices:
light.office 'Office Light' = on;80%
fan.office 'Office fan' = off
light.kitchen 'Kitchen Light' = on;80%;red
light.bedroom 'Bedroom Light' = off

The model's output includes a response for the user and an optional code block to invoke Home Assistant "services". The output format for function calling is as follows:

turning on the kitchen lights for you now
```homeassistant
{ "service": "light.turn_on", "target_device": "light.kitchen" }
```

The model can also handle basic instruct and QA tasks due to instruction fine - tuning in the base model. For example:

user if mary is 7 years old, and I am 3 years older than her. how old am I?
assistant If Mary is 7 years old, then you are 10 years old (7+3=10).

✨ Features

• Accuracy Improvement: V3 uses a new base model (StableLM), resulting in significant accuracy increases.
• Multi - Support: It offers basic multi - personality support, basic multi - language support (for English, German, Spanish, and French), and support for more Home Assistant entity types (vacuum, timer, and todo).
• Low - Resource Compatibility: Quantized using Lama.cpp, it can run in low - resource environments like Raspberry Pis commonly used in Home Assistant installations.

🔧 Technical Details

Training

The model was trained as a LoRA on an RTX 3090 (24GB). The LoRA has a rank of 64, an alpha of 128, and targets the up_proj,down_proj,q_proj,v_proj,o_proj modules. The full model is merged at the end.

Evaluation

This model achieves a 97.11% score for JSON function calling accuracy on the test dataset.

📦 Datasets

📄 License

This model is a fine - tuned version of the Stability AI StableLM model series, which is licensed under the STABILITY AI NON - COMMERCIAL RESEARCH COMMUNITY LICENSE AGREEMENT. Therefore, this model is released under the same non - commercial license and is for non - commercial use only.

⚠️ Important Note

The base models do not have native multi - language support but use a tokenizer that can handle non - English languages better than Phi - 2. It has been verified to work in German, Spanish, and French on some random examples where the request is an English request processed via Google Translate.