app.py

import gradio as gr
from transformers import AutoProcessor, AutoModelForVision2Seq
import torch

# Load model
print("Loading Afri-Aya model...")

processor = AutoProcessor.from_pretrained("Bronsn/afri-aya-gemma-3-4b-vision")
model = AutoModelForVision2Seq.from_pretrained(
    "Bronsn/afri-aya-gemma-3-4b-vision",
    torch_dtype=torch.float16,
    device_map="auto"
)

print("Model loaded successfully!")

def predict(image, question):
    if image is None:
        return "Please upload an image"
    
    # Prepare input
    inputs = processor(images=image, text=question, return_tensors="pt").to(model.device)
    
    # Generate
    output_ids = model.generate(**inputs, max_new_tokens=100)
    
    # Decode
    answer = processor.decode(output_ids[0], skip_special_tokens=True)
    
    return answer

# Create interface
demo = gr.Interface(
    fn=predict,
    inputs=[
        gr.Image(type="pil", label="Upload African Cultural Image"),
        gr.Textbox(label="Ask a question", placeholder="What is shown in this image?")
    ],
    outputs=gr.Textbox(label="Answer"),
    title="🌍 Afri-Aya Vision Model",
    description="Ask questions about African cultural images in multiple languages!"
)

demo.launch()