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 for usage terms.