handler.py

from typing import Dict, Any, List
import torch
import logging
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

class EndpointHandler():
    def __init__(self, path=""):
        self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
        try:
            logger.info(f"Loading model and tokenizer from path: {path}")
            self.model = AutoModelForSeq2SeqLM.from_pretrained(f"{path}").to(self.device)
            self.tokenizer = AutoTokenizer.from_pretrained(f"{path}")
        except Exception as e:
            logger.error(f"Error loading model or tokenizer from path {path}: {e}")
            # Handle error (e.g., exit or set model/tokenizer to None)

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        if self.model is None or self.tokenizer is None:
            error_message = "Model or tokenizer not properly initialized"
            logger.error(error_message)
            return [{"error": error_message}]
        inputs = data.get("inputs")
        if not inputs:
            return [{"error": "No inputs provided"}]

        tokenized_input = self.tokenizer(inputs, return_tensors="pt")
        input_ids,attention_masks = tokenized_input["input_ids"].to(self.device), tokenized_input["attention_mask"].to(self.device)  # Move input tensors to the same device as model

        summary_ids = self.model.generate(input_ids=input_ids, attention_mask=attention_masks,)

        summary_text = self.tokenizer.decode(summary_ids[0], skip_special_tokens=True,  clean_up_tokenization_spaces=True)
        print('good')
        return [{"summary": summary_text}]