app.py

from transformers import AutoProcessor, PaliGemmaForConditionalGeneration
from PIL import Image
import torch
import gradio as gr

# Load model and processor
processor = AutoProcessor.from_pretrained("google/paligemma-3b")
model = PaliGemmaForConditionalGeneration.from_pretrained(
    "google/paligemma-3b", 
    torch_dtype=torch.float16, 
    device_map="auto"
)

def ecommerce_assistant(image, question):
    prompt = f"Question: {question}\nAnswer:"
    inputs = processor(text=prompt, images=image, return_tensors="pt").to("cuda", torch.float16)
    outputs = model.generate(**inputs, max_new_tokens=50)
    return processor.batch_decode(outputs, skip_special_tokens=True)[0]

demo = gr.Interface(
    fn=ecommerce_assistant,
    inputs=["image", "text"],
    outputs="text",
    title="🛍️ E-commerce Visual Assistant",
    description="Upload a product photo and ask questions like 'What brand is this?' or 'Can I buy it online?'"
)

demo.launch()