handler.py

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

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

class EndpointHandler():
    def __init__(self, path=""):
        self.processor = AutoProcessor.from_pretrained(path)
        self.model = Qwen2VLForConditionalGeneration.from_pretrained(
            path, device_map="auto"
        )
        self.model.to(device)

    def __call__(self, data: Any) -> Dict[str, Any]:
        inputs = data.pop("inputs", data)
        image_input = inputs.get('image')
        text_input = inputs.get('text', "Describe this image.")

        if not image_input:
            return {"error": "No image provided."}

        try:
            if image_input.startswith('http'):
                response = requests.get(image_input, stream=True)
                if response.status_code == 200:
                    image = Image.open(response.raw).convert('RGB')
                else:
                    return {"error": f"Failed to fetch image. Status code: {response.status_code}"}
            else:
                image_data = base64.b64decode(image_input)
                image = Image.open(io.BytesIO(image_data)).convert('RGB')
        except Exception as e:
            return {"error": f"Failed to process the image. Details: {str(e)}"}

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

            text_prompt = self.processor.apply_chat_template(
                conversation, add_generation_prompt=True
            )

            inputs = self.processor(
                text=[text_prompt],
                images=[image],
                padding=True,
                return_tensors="pt",
            )

            inputs = inputs.to(device)

            output_ids = self.model.generate(
                **inputs, max_new_tokens=128
            )

            generated_ids = [
                output_id[len(input_id):] for input_id, output_id in zip(inputs.input_ids, output_ids)
            ]

            output_text = self.processor.batch_decode(
                generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True
            )[0]

            return {"generated_text": output_text}

        except Exception as e:
            return {"error": f"Failed during generation. Details: {str(e)}"}