from transformers import AutoModelForSequenceClassification, AutoTokenizer

def load_model_and_tokenizer():
    """
    Load the fine-tuned XLM-RoBERTa model and tokenizer.
    Returns the model and tokenizer for use in classification.
    """
    try:
        model_name = "JanviMl/xlm-roberta-toxic-classifier-capstone"  # Replace with your model repo ID
        # If the model is local: model_name = "./model"
        
        model = AutoModelForSequenceClassification.from_pretrained(model_name)
        tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)  # Use slow tokenizer
        return model, tokenizer
    except Exception as e:
        raise Exception(f"Error loading model or tokenizer: {str(e)}")

# Load the model and tokenizer once at startup
model, tokenizer = load_model_and_tokenizer()