tools/polite_guard.py

from typing import Any, Optional
from smolagents.tools import Tool
from transformers import pipeline


class PoliteGuardTool(Tool):
    """
        Takes the input text from users and then evaluates it against Polite Guard to return specific information
        about whether the content is polite. 
        
        Args:
            input_text: Text that the user inputs into the agent and should then be evaluated. 
            
        Returns:
            A classification label about whether the content is polite, somewhat polite, neutral or impolite.
            
    """

    name = "polite_guard"
    description = "Uses Polite guard to classify input text from polite to impolite"
    inputs = {'input_text': {'type': 'any', 'description': 'Enter text for assessing whether it is respectful'}}
    output_type = "any"

    def forward(self, input_text: Any) -> Any:
        self.label, self.score = self.ask_polite_guard(input_text)
        print(f"forward sets the following: {self.label } and {self.score}")
        return self.label 
           
    def __init__(self, *args, **kwargs):
        self.is_initialized = False
        self.label = None
        self.score = None
        
    def ask_polite_guard(self, input_text: str) -> tuple[str, float]:
        """ 
    
        Args:
            input_text: The text to classify.
    
        Returns:
            tuple: with classification label and the score 
        """
        try:
            classifier = pipeline("text-classification", "Intel/polite-guard")
            result = classifier(input_text)
            print(f"return {str(result)}")
            output = result[0]
            return output['label'], output['score']
            
        except Exception as e:
            return f"Error fetching classification for text '{input_text}': {str(e)}"