from transformers import LlavaNextProcessor, LlavaNextForConditionalGeneration
import torch
from PIL import Image

class EndpointHandler():
    def __init__(self, path=""):
        disable_torch_init()
        self.processor = LlavaNextProcessor.from_pretrained(path, use_fast=False)
        self.model = LlavaNextForConditionalGeneration.from_pretrained(
            path, 
            torch_dtype=torch.float16, 
            low_cpu_mem_usage=True,
            load_in_4bit=True
        )
        self.model.to("cuda:0")

    def __call__(self, data):
        image_encoded = data.pop("inputs", data)
        prompt = data["text"]

        image = self.decode_base64_image(image_encoded)
        if image.mode != "RGB":
            image = image.convert("RGB")

        inputs = self.processor(prompt, image, return_tensors="pt").to("cuda:0")

        # autoregressively complete prompt
        output = self.model.generate(**inputs, max_new_tokens=500)

        return processor.decode(output[0], skip_special_tokens=True)