from PIL import Image
from transformers import AutoProcessor, BlipForQuestionAnswering
import torch


class blip:
    device = "cuda" if torch.cuda.is_available() else "cpu"
    
    
    def __init__(self, model_pretrain:str = "Salesforce/blip-vqa-base"):
        self.processor = AutoProcessor.from_pretrained(model_pretrain)
        self.model = BlipForQuestionAnswering.from_pretrained(
            model_pretrain, device_map={"": 0}, torch_dtype=torch.float16
            )
        
    def image_captioning(self, image: Image.Image) -> str:
        
        text = "What are the features of this picture??"
        inputs = self.processor(images=image, text=text, return_tensors="pt").to(self.device, torch.float16)
        outputs = self.model.generate(**inputs)
        
        return self.processor.decode(outputs[0], skip_special_tokens=True)
    
    def visual_question_answering(self, image: Image.Image, prompt: str) -> str:
        inputs = self.processor(images=image, text=prompt, return_tensors="pt").to(self.device, torch.float16)

        generated_ids = self.model.generate(**inputs)
        generated_text = self.processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
        
        return generated_text