import gradio as gr
from PIL import Image
import torch
from transformers import BlipProcessor, BlipForConditionalGeneration

processor = BlipProcessor.from_pretrained("zeddotes/blip-computer-thoughts")
model = BlipForConditionalGeneration.from_pretrained("zeddotes/blip-computer-thoughts")

def caption_image(image):
    # image is a PIL Image from Gradio
    # Convert to model inputs
    inputs = processor(images=image, return_tensors="pt")
    with torch.no_grad():
        # Generate text from the model
        generated_ids = model.generate(**inputs, max_length=50)
    caption = processor.decode(generated_ids[0], skip_special_tokens=True)
    return caption

demo = gr.Interface(
    fn=caption_image,
    inputs=gr.Image(type="pil"),
    outputs="text",
    title="My Fine-Tuned BLIP Model"
)

if __name__ == "__main__":
    demo.launch()