from typing import Dict, Any
import torch
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
from PIL import Image
import io
import base64
import requests
from qwen_vl_utils import process_vision_info

class EndpointHandler():
    def __init__(self, path=""):
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        self.model = Qwen2VLForConditionalGeneration.from_pretrained(
            path,
            torch_dtype="auto",
            device_map="auto"
        ).to(self.device)
        
        self.processor = AutoProcessor.from_pretrained(path)
        
        # Optionally, adjust min_pixels and max_pixels if needed
        # min_pixels = 256*28*28
        # max_pixels = 1280*28*28
        # self.processor = AutoProcessor.from_pretrained(path, min_pixels=min_pixels, max_pixels=max_pixels)

    def __call__(self, data: Any) -> Dict[str, Any]:
        """
        Args:
            data (Any): The input data, which can be:
                - Binary image data in the request body.
                - A dictionary with 'image' and 'text' keys:
                    - 'image': Base64-encoded image string or image URL.
                    - 'text': The text prompt.

        Returns:
            Dict[str, Any]: The generated text output from the model.
        """
        default_prompt = "Describe this image."
        
        if isinstance(data, (bytes, bytearray)):
            image = Image.open(io.BytesIO(data)).convert('RGB')
            text_input = default_prompt
        elif isinstance(data, dict):
            image_input = data.get('image', None)
            text_input = data.get('text', default_prompt)
            if image_input is None:
                return {"error": "No image provided."}
            if image_input.startswith('http'):
                response = requests.get(image_input)
                image = Image.open(io.BytesIO(response.content)).convert('RGB')
            else:
                image_data = base64.b64decode(image_input)
                image = Image.open(io.BytesIO(image_data)).convert('RGB')
        else:
            return {"error": "Invalid input data. Expected binary image data or a dictionary with 'image' key."}

        messages = [
            {
                "role": "user",
                "content": [
                    {
                        "type": "image",
                        "image": image,
                    },
                    {"type": "text", "text": text_input},
                ],
            }
        ]

        text = self.processor.apply_chat_template(
            messages, tokenize=False, add_generation_prompt=True
        )
        image_inputs, video_inputs = process_vision_info(messages)
        inputs = self.processor(
            text=[text],
            images=image_inputs,
            videos=video_inputs,
            padding=True,
            return_tensors="pt",
        )
        inputs = inputs.to(self.device)

        generated_ids = self.model.generate(**inputs, max_new_tokens=128)
        generated_ids_trimmed = [
            out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
        ]
        output_text = self.processor.batch_decode(
            generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
        )

        return {"generated_text": output_text[0]}