handler.py

import logging
from typing import Any, Dict

import torch.cuda
from peft import PeftConfig, PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

LOGGER = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)
device = "cuda" if torch.cuda.is_available() else "cpu"


class EndpointHandler():
    def __init__(self, path=""):
        config = PeftConfig.from_pretrained(path)

        model = AutoModelForCausalLM.from_pretrained(
            config.base_model_name_or_path,
            load_in_8bit=True,
            trust_remote_code=True,
            device_map="auto"
        )
        
        self.tokenizer = AutoTokenizer.from_pretrained(
            config.base_model_name_or_path, trust_remote_code=True)
        self.tokenizer.pad_token = self.tokenizer.eos_token
        
        # Load the Lora model
        self.model = PeftModel.from_pretrained(model, path, torch_dtype=model.dtype)
        self.model.eos_token_id = self.tokenizer.eos_token_id

    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
        """
        Args:
            data (Dict): The payload with the text prompt and generation parameters.
        """
        LOGGER.info(f"Received data: {data}")
        # Get inputs
        prompt = data.pop("inputs", None)
        parameters = data.pop("parameters", None)
        if prompt is None:
            raise ValueError("Missing prompt.")
        # Preprocess
        input_ids = self.tokenizer(
            prompt, return_tensors="pt").input_ids.to(device)
        # Forward
        LOGGER.info(f"Start generation.")
        if parameters is not None:
            output = self.model.generate(input_ids=input_ids, **parameters)
        else:
            output = self.model.generate(input_ids=input_ids)
        # Postprocess
        prediction = self.tokenizer.decode(output[0])
        LOGGER.info(f"Generated text: {prediction}")
        return {"generated_text": prediction}