handler.py

import os
from huggingface_hub import login
from peft import PeftModel
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from typing import Dict, List, Any


class EndpointHandler:
    def __init__(self, path=""):
        login(token=os.environ["HF_ACCESS_TOKEN"])
        
        # load model and tokenizer from path
        self.tokenizer = AutoTokenizer.from_pretrained(path)

        bnb_config = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_quant_type='nf4',
            bnb_4bit_compute_dtype=torch.bfloat16,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_storage=torch.bfloat16)
        base_model = AutoModelForCausalLM.from_pretrained(
            "meta-llama/Llama-3.2-11B-Vision-Instruct",
            quantization_config=bnb_config,
            device_map="auto",
            torch_dtype=torch.bfloat16)
        
        self.model = PeftModel.from_pretrained(base_model, path)
        self.device = "cuda" if torch.cuda.is_available() else "cpu"

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:

        print(data)
        print(type(data))
        
        if isinstance(data, dict) and "inputs" in data:
            inputs = data["inputs"]
        else:
            return {"error": "Invalid input format. Expecting {'inputs': 'your text'}"}
            
        inputs = self.tokenizer(
            inputs, 
            return_tensors="pt", 
            padding=True, 
            truncation=True, 
            max_length=700).to(self.device)

        with torch.no_grad():
            outputs = self.model.generate(
                input_ids=inputs["input_ids"],
                attention_mask=inputs["attention_mask"],
                max_new_tokens=700,
                temperature=0.01)

        # postprocess the prediction
        prediction = self.tokenizer.decode(outputs[0], skip_special_tokens=True)

        return [{"generated_text": prediction}]