handler.py

import torch
from PIL import Image
import requests
from io import BytesIO
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
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")
        
        # Load model and processor
        self.model = Qwen2VLForConditionalGeneration.from_pretrained(
            path, torch_dtype=torch.float16 if self.device.type == "cuda" else torch.float32, device_map="auto"
        ).to(self.device)
        self.processor = AutoProcessor.from_pretrained(path)

    def __call__(self, data):
        # Extract image and text from the input data
        image_url = data.get("inputs", {}).get("image", "")
        text_prompt = data.get("inputs", {}).get("text", "")

        if not image_url or not text_prompt:
            return {"error": "Both 'image' and 'text' must be provided in the input data."}

        # Download and process the image
        try:
            response = requests.get(image_url)
            response.raise_for_status()
            image = Image.open(BytesIO(response.content)).convert("RGB")
        except Exception as e:
            return {"error": f"Failed to load image from URL: {e}"}

        # Prepare the input in the format expected by the model
        messages = [
            {
                "role": "user",
                "content": [
                    {"type": "image", "image": image},
                    {"type": "text", "text": text_prompt},
                ],
            }
        ]

        # Process the 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",
        )

        # Move inputs to the appropriate device
        inputs = {k: v.to(self.device) for k, v in inputs.items()}

        # Generate output
        with torch.no_grad():
            output_ids = self.model.generate(**inputs, max_new_tokens=128)

        # Decode the output
        output_text = self.processor.batch_decode(
            output_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
        )[0]

        return {"generated_text": output_text}