---
language:
- en
- id
license: apache-2.0
tags:
- iot
- edge-ai
- raspberry-pi
- ollama
- gguf
- lora
- qwen2
- plutoclaw
base_model: Qwen/Qwen2.5-1.5B-Instruct
model-index:
- name: PlutoEdge-1.5B
  results: []
---

# PlutoEdge-1.5B

**PlutoEdge-1.5B** is a domain-specific LLM fine-tuned for IoT edge automation, running on Raspberry Pi via Ollama. It powers [PlutoClaw](https://github.com/plutoedge-dev/plutoclaw) — an open-source Edge AI orchestrator for physical hardware control.

> Runs fully offline on Raspberry Pi CPU. No GPU, no cloud, no API keys.

## Model Details

| Property | Value |
|---|---|
| **Base model** | Qwen2.5-1.5B-Instruct |
| **Fine-tuning** | MLX LoRA (rank=16, 1500 iters) |
| **Format** | GGUF Q4_K_M |
| **Size** | ~940 MB |
| **Raspberry Pi inference** | ~37s / response (CPU) |
| **Context window** | 1024 tokens (Pi) / 2048 tokens (Mac) |
| **Training samples** | 759 (synthetic + acon96/Home-Assistant-Requests) |
| **Language** | English (Bahasa Indonesia input supported via normalization) |

## What It Does

PlutoEdge understands IoT control commands and responds with structured `PLUTO_ACTION` JSON that PlutoClaw executes on GPIO hardware:

```
User:  "Turn on the ventilation fan"
Pluto: "Turning on the ventilation fan now."
       PLUTO_ACTION: {"type": "actuator_trigger", "params": {"id": "relay1", "action": "on"}}
```

```
User:  "Worker detected without hard hat"
Pluto: "PPE violation detected. Sounding alert buzzer."
       PLUTO_ACTION: {"type": "multi_trigger", "params": [{"id": "buzzer1", "action": "pulse"}, {"id": "led1", "action": "on"}]}
```

## Training Domains

| Domain | Samples | Skills |
|---|---|---|
| Smart Home | 520 | relay control, automation, flood/fire detection |
| Knowledge Q&A | 66 | PlutoClaw platform, skill selection, setup |
| Warehouse | 41 | ppe_guard, intrusion, forklift_guard |
| Sustainability | 34 | solar/grid, carbon footprint, water monitoring |
| Poultry Farming | 33 | coop_monitor, sick_animal, animal_count |
| Industrial | 30 | predictive_maintenance, quality_control |
| Agriculture | 27 | irrigation_control, crop_monitor |

## PLUTO_ACTION Format

```json
// Single device
PLUTO_ACTION: {"type": "actuator_trigger", "params": {"id": "relay1", "action": "on"}}

// Multiple devices simultaneously
PLUTO_ACTION: {"type": "multi_trigger", "params": [
  {"id": "relay1", "action": "off"},
  {"id": "buzzer1", "action": "on"},
  {"id": "led1", "action": "on"}
]}
```

## Quickstart with Ollama

**Option 1 — Pull directly from HuggingFace:**
```bash
# Install Ollama on Raspberry Pi
curl -fsSL https://ollama.ai/install.sh | sh

# Pull and run PlutoEdge
ollama pull hf.co/plutoedge/PlutoEdge-1.5B
ollama run hf.co/plutoedge/PlutoEdge-1.5B
```

**Option 2 — Build from PlutoClaw repo (recommended for full GPIO automation):**
```bash
# Install Ollama on Raspberry Pi
curl -fsSL https://ollama.ai/install.sh | sh

# Clone PlutoClaw and register PlutoEdge locally
git clone https://github.com/plutoedge-dev/plutoclaw.git
cd plutoclaw/models/PlutoEdge-1.5B-v4
ollama create plutoedge -f Modelfile
```

Or use with [PlutoClaw](https://github.com/plutoedge-dev/plutoclaw) for full GPIO automation:

```bash
git clone https://github.com/plutoedge-dev/plutoclaw.git
cd plutoclaw
pip install -r requirements.txt
# Edit config.yaml, then:
python3 main.py
```

## Files

| File | Description |
|---|---|
| `PlutoEdge-1.5B-v4-Q4_K_M.gguf` | Quantized model for Raspberry Pi (940 MB) |
| `PlutoEdge-1.5B-v4-F16.gguf` | Full precision GGUF (3.1 GB) |
| `Modelfile` | Ollama Modelfile with system prompt |

## License

Apache 2.0 — same as base model (Qwen2.5-1.5B-Instruct by Alibaba Cloud).

---

Built by [Plutobot AI](https://plutobot.ai) · Jakarta, Indonesia