from transformers import AutoProcessor, AutoModelForCausalLM
from huggingface_hub import hf_hub_download
from PIL import Image
import gradio as gr
import torch


processor = AutoProcessor.from_pretrained("microsoft/git-large-vqav2")
model = AutoModelForCausalLM.from_pretrained("microsoft/git-large-vqav2")


def vqa(image, question):
    inp = Image.fromarray(image.astype('uint8'), 'RGB')
    pixel_values = processor(images=inp, return_tensors="pt").pixel_values

    input_ids = processor(text=question, add_special_tokens=False).input_ids
    input_ids = [processor.tokenizer.cls_token_id] + input_ids
    input_ids = torch.tensor(input_ids).unsqueeze(0)

    generated_ids = model.generate(pixel_values=pixel_values, input_ids=input_ids, max_length=50)
    return processor.batch_decode(generated_ids, skip_special_tokens=True)

demo = gr.Interface(fn=vqa,
                    inputs=['image', 'text'],
                    outputs='text',
                    title = "vqa",
                    examples = [['soccer.jpg', 'how many people in the picture?']])


if __name__ == "__main__":
    demo.queue(concurrency_count=3).launch(server_name = "0.0.0.0", server_port = 7024)