from typing import Dict, List, Any
from PIL import Image
from io import BytesIO
import base64
import torch
import open_clip

class EndpointHandler():
    def __init__(self, path=""):
        self.model, self.preprocess, _ = open_clip.create_model_and_transforms('hf-hub:laion/CLIP-ViT-H-14-laion2B-s32B-b79K')
        self.tokenizer = open_clip.get_tokenizer('hf-hub:laion/CLIP-ViT-H-14-laion2B-s32B-b79K')

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        image_base64 = data.get("inputs", None)
        parameters = data.get("parameters", None)
        if image_base64 is None or parameters is None:
            raise ValueError("Input data or parameters not provided")

        candidate_labels = parameters.get("candidate_labels", None)
        if candidate_labels is None:
            raise ValueError("Candidate labels not provided")

        image = Image.open(BytesIO(base64.b64decode(image_base64)))
        image = self.preprocess(image).unsqueeze(0)
        text = self.tokenizer(candidate_labels)

        with torch.no_grad():
            image_features = self.model.encode_image(image)
            text_features = self.model.encode_text(text)
            image_features /= image_features.norm(dim=-1, keepdim=True)
            text_features /= text_features.norm(dim=-1, keepdim=True)

            text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)

        results = [{"label": label, "score": score.item()} for label, score in zip(candidate_labels, text_probs[0])]
        return results