import torch
import requests
from PIL import Image
from transformers import ViTFeatureExtractor, AutoTokenizer, VisionEncoderDecoderModel
import gradio as gr


loc = "ydshieh/vit-gpt2-coco-en"

feature_extractor = ViTFeatureExtractor.from_pretrained(loc)
tokenizer = AutoTokenizer.from_pretrained(loc)
model = VisionEncoderDecoderModel.from_pretrained(loc)
model.eval()


def predict(image):
    pixel_values = feature_extractor(images=image, return_tensors="pt").pixel_values

    with torch.no_grad():
        output_ids = model.generate(pixel_values, max_length=16, num_beams=4, return_dict_in_generate=True).sequences

    preds = tokenizer.batch_decode(output_ids, skip_special_tokens=True)

    total_caption = ""
    for pred in preds:
        total_caption = total_caption + pred.strip()
        total_caption = total_caption + "\r\n"

    return total_caption

demo = gr.Interface(fn=predict,
                    inputs='image',
                    outputs='text',
                    title = "image2text",
                    examples = ['soccer.jpg'])


if __name__ == "__main__":
    demo.queue(concurrency_count=3).launch(server_name = "0.0.0.0", server_port = 7019)