handler.py

from typing import Dict, Any
from PIL import Image    
import requests
import flash-attn
import torch
import numpy as np
from transformers import AutoProcessor, LlavaForConditionalGeneration, BitsAndBytesConfig


class EndpointHandler():
    def __init__(self, path=""):
        model_id = path
        self.model = LlavaForConditionalGeneration.from_pretrained(
            model_id, 
            torch_dtype=torch.float16, 
            low_cpu_mem_usage=True,
            use_flash_attention_2=True
        )
        self.processor = AutoProcessor.from_pretrained(model_id)

    def __call__(self, data: Dict[list, Any]):
        parameters = data.pop("inputs", data)
        givenprompt = data.pop("prompt", data)
        outputs = []
        print(parameters)
        prompt = f"USER: <image>\n{givenprompt}?\nASSISTANT:"
        for link in parameters:
            try:
                # Fetch image from URL
                response = requests.get(link, stream=True)
                response.raise_for_status()  # Raise an exception for 4xx or 5xx status codes
                raw_image = Image.open(response.raw)
                # Process image and generate output
                inputs = self.processor(prompt, raw_image, return_tensors='pt').to(0, torch.float16)
                output = self.model.generate(**inputs, max_new_tokens=200, do_sample=False)
                readable = self.processor.decode(output[0][2:], skip_special_tokens=True)
                outputs.append(readable)
            except Exception as e:
                # Handle any exceptions and log the error
                outputs.append(f"Error processing image from {link}: {str(e)}")
        
        return outputs