---
license: mit
base_model: google/gemma-4-e4b-it
library_name: peft
tags:
  - gemma4
  - solar-energy
  - community-solar
  - function-calling
  - multimodal
  - lora
  - unsloth
  - qlora
  - ollama
  - edge-ai
  - energy
  - sustainability
  - hackathon
datasets:
  - solarhive-community-solar-1k
language:
  - en
pipeline_tag: image-text-to-text
model-index:
  - name: SolarHive-E4B-LoRA
    results:
      - task:
          type: question-answering
          name: Domain Q&A
        metrics:
          - name: Accuracy
            type: accuracy
            value: 1.0
            verified: false
      - task:
          type: text-generation
          name: Tool Calling
        metrics:
          - name: Accuracy
            type: accuracy
            value: 1.0
            verified: false
---

![SolarHive](SolarHive_HeaderImage_1920x1080_HFModelCard.png)

# SolarHive E4B Ollama — Edge Solar Energy Intelligence

**QLoRA fine-tuned Gemma 4 E4B (8B)**, merged to 16-bit safetensors for **edge deployment via Ollama**. Community energy data never leaves the neighborhood.

This repository contains **merged safetensors** (base model + LoRA adapters baked together). Ollama imports safetensors directly — no GGUF conversion needed.

Built for the [Gemma 4 Good Hackathon](https://kaggle.com/competitions/gemma-4-good-hackathon) (Google DeepMind x Kaggle).

| | |
|---|---|
| **Base Model** | [google/gemma-4-e4b-it](https://kaggle.com/models/google/gemma-4) |
| **Architecture** | Dense + PLE — 8B total, 4.5B effective |
| **Fine-Tuning** | QLoRA via [Unsloth](https://unsloth.ai) |
| **Training Data** | 1,029 community solar energy examples |
| **Converged Loss** | **0.952** |
| **Benchmark** | 7/8 (5/5 domain Q&A + 2/3 tool calling) |
| **Training Time** | 282 seconds (~4.7 minutes) |
| **Compute** | Google Colab Pro |
| **License** | MIT (adapters) / Gemma Terms (base model) |

---

## Model Overview

SolarHive E4B is the **edge companion** to [SolarHive 26B A4B](https://huggingface.co/Truthseeker87/solarhive-26b-a4b-lora). While the 26B model powers cloud inference with full multimodal VQA, the E4B model is optimized for local deployment via Ollama on consumer hardware.

**Privacy-first:** Running Gemma 4 locally means community energy data never leaves the neighborhood. No cloud dependency, no internet requirement, no data privacy concerns. A village in rural India, a suburb in Michigan, and a coastal town recovering from a hurricane all get the same intelligence.

This repository contains **LoRA adapters only** — you need the base Gemma 4 E4B model to use them.

---

## Training Details

| Parameter | Value |
|-----------|-------|
| Method | QLoRA via Unsloth `FastVisionModel` |
| LoRA rank | 16 |
| LoRA alpha | 16 |
| LoRA dropout | 0 |
| Target modules | All linear layers |
| Learning rate | 2e-4 |
| Optimizer | AdamW 8-bit |
| Warmup steps | 5 |
| Epochs | 3 |
| Max sequence length | 2048 |
| Precision | BF16 |
| Seed | 3407 |
| Trainable parameters | 41.2M / 8.0B (0.51%) |

### Training Loss

| Metric | Value |
|--------|-------|
| Converged loss (last 20 steps) | **0.952** |
| Final step loss | 1.088 |
| Minimum loss | 0.455 |
| Total steps | 195 |
| Training time | 282 seconds |

### Training Data

Same 1,029 examples as the 26B A4B model:
- 413 hand-crafted examples spanning 15+ US cities and 9 energy domains
- 501 API-grounded examples from live Open-Meteo, PVWatts, OWM, and EIA data
- 100 tool-calling examples (50 with tools, 50 without)

See the [SolarHive Dataset](https://huggingface.co/datasets/Truthseeker87/solarhive-community-solar-1k) for full documentation.

### Hardware

- **GPU:** NVIDIA RTX PRO 6000 Blackwell Server Edition (96 GB GDDR7)
- **Platform:** Google Colab G4 VM
- **Platform:** Google Colab Pro

---

## Benchmark Results

### Domain Q&A (5/5)

| Question | Result |
|----------|--------|
| Solar production when humidity exceeds 80%? | Correct |
| Battery SOC threshold for grid export? | Correct |
| Home #3 underperforming 22% — diagnostic checklist? | Correct |
| Winter snow on panels — prioritize actions? | Correct |
| Grid frequency 59.8 Hz — microgrid implications? | Correct |

### Tool Calling (2/3)

| Question | Expected | Called | Status |
|----------|----------|-------|--------|
| Current battery state? | `get_battery_state` | `get_battery_status` | Fail |
| Solar production in Seattle? | `get_solar_production` or `get_weather` | `get_solar_production` | Pass |
| General panel maintenance tips? | None | None | Pass |

---

## How to Use

### Loading with Unsloth

```python
from unsloth import FastVisionModel
import torch

model, processor = FastVisionModel.from_pretrained(
    model_name="google/gemma-4-e4b-it",
    adapter_name="Truthseeker87/solarhive-e4b-lora",  # This repo
    dtype=torch.bfloat16,
    device_map="auto",
)
FastVisionModel.for_inference(model)
```

### Edge Deployment via Ollama

```bash
# After converting to GGUF (see solarhive_finetune.py Cell 13):
ollama create solarhive -f Modelfile
ollama run solarhive "What's the best time to run my dishwasher today?"
```

---

## Companion Repositories

| Model | Repository | Purpose |
|-------|-----------|---------|
| **SolarHive 26B A4B LoRA** | [solarhive-26b-a4b-lora](https://huggingface.co/Truthseeker87/solarhive-26b-a4b-lora) | Cloud inference with full multimodal + function calling |
| **SolarHive E4B Ollama** | This repo | Edge deployment via Ollama (merged safetensors) |
| **SolarHive Dataset** | [solarhive-community-solar-1k](https://huggingface.co/datasets/Truthseeker87/solarhive-community-solar-1k) | 1,029 training examples |
| **GitHub** | [the-gemma4-good-hackathon-solarhive](https://github.com/youshen-lim/the-gemma4-good-hackathon-solarhive) | Full source code, notebooks, data principles |

---

## Links

- **GitHub:** [youshen-lim/the-gemma4-good-hackathon-solarhive](https://github.com/youshen-lim/the-gemma4-good-hackathon-solarhive)
- **Kaggle:** [The Gemma 4 Good Hackathon](https://kaggle.com/competitions/gemma-4-good-hackathon)

*Built with Gemma 4 in Ann Arbor, Michigan. April 2026.*

*Gemma is a trademark of Google LLC.*