---
library_name: peft
base_model: google/gemma-3-4b-it
tags:
- vision
- image-classification
- beans
- plant-disease
- gemma-3
- lora
- fine-tuned
license: gemma
---

# Gemma-3-4B Fine-tuned for Bean Disease Classification

This model is a fine-tuned version of [google/gemma-3-4b-it](https://huggingface.co/google/gemma-3-4b-it) for classifying bean plant diseases.

## Model Description

- **Base Model:** Gemma-3-4B-IT (Vision)
- **Fine-tuning Method:** LoRA (r=8, alpha=16)
- **Dataset:** [beans](https://huggingface.co/datasets/beans) (100 samples)
- **Task:** Image captioning / disease classification
- **Final Validation Loss:** 0.001 (excellent!)

## Classes

1. Healthy bean plant
2. Angular leaf spot disease
3. Bean rust disease

## Usage

```python
from transformers import AutoProcessor, Gemma3ForConditionalGeneration
from peft import PeftModel
from PIL import Image
import torch

# Load base model
base_model = Gemma3ForConditionalGeneration.from_pretrained(
    "google/gemma-3-4b-it",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "younaice/gemma3-4b-bean-captioning")
processor = AutoProcessor.from_pretrained("younaice/gemma3-4b-bean-captioning")

# Prepare input
image = Image.open("bean_plant.jpg")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image"},
            {"type": "text", "text": "Describe this plant image."}
        ]
    }
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=text, images=image, return_tensors="pt").to(model.device)

# Generate
outputs = model.generate(**inputs, max_new_tokens=50, do_sample=False)
response = processor.decode(outputs[0], skip_special_tokens=True)
print(response)
```

## Training Details

- **Epochs:** 10
- **Batch Size:** 1 (effective: 4 with gradient accumulation)
- **Learning Rate:** 5e-5
- **Precision:** FP16
- **Hardware:** NVIDIA T4 GPU
- **Training Time:** ~25 minutes
- **Max Sequence Length:** 512 tokens

## Performance

- **Final Training Loss:** 0.69
- **Final Validation Loss:** 0.001
- **Accuracy:** Very high (based on validation loss)

## Limitations

- Trained on 100 images for demonstration purposes
- Best suited for the 3 specific bean disease types in the training data
- May not generalize to other bean varieties or diseases
- Should be validated on real-world data before production use

## Citation

If you use this model, please cite:

```bibtex
@misc{gemma3-bean-captioning,
  author = {younaice},
  title = {Gemma-3-4B Fine-tuned for Bean Disease Classification},
  year = {2024},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/younaice/gemma3-4b-bean-captioning}}
}
```

## License

This model inherits the Gemma license from the base model. Please refer to the [Gemma license](https://ai.google.dev/gemma/terms) for usage terms.