handler.py

from typing import Dict, Any
import torch
from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
from PIL import Image
import requests
from io import BytesIO

# Check for GPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

class EndpointHandler:
    def __init__(self, path: str = "morthens/qwen2-vl-inference"):
        # Load the processor and model
        self.processor = AutoProcessor.from_pretrained(path)
        self.model = Qwen2VLForConditionalGeneration.from_pretrained(
            path,
            torch_dtype="auto",
            device_map="auto"
        )
        # Move the model to the appropriate device
        self.model.to(device)

    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
        # Extract the input data
        image_url = data.get("image_url", "")
        text = data.get("text", "")

        # Load the image from the URL
        try:
            response = requests.get(image_url)
            response.raise_for_status()
            image = Image.open(BytesIO(response.content))
        except Exception as e:
            return {"error": f"Failed to fetch or process image: {str(e)}"}

        # Preprocess the input
        inputs = self.processor(
            text=[text],
            images=[image],
            padding=True,
            return_tensors="pt"
        )

        # Move inputs to the correct device
        inputs = {key: value.to(device) for key, value in inputs.items()}

        # Perform inference
        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=True
        )[0]

        # Return the raw prediction
        return {"prediction": output_text}