app.py

def respond(message, history, max_tokens, temperature, top_p):
    """Generate a response while preventing unnecessary repetition."""
    
    # Ensure the message isn't directly repeated
    if history and message.strip() == history[-1][0].strip():
        return "Tafadhali uliza swali tofauti."  # Prevent direct repetition
    
    response = client.text_generation(
        message.strip(),
        max_new_tokens=max_tokens,  
        temperature=temperature,
        top_p=top_p,
    ).strip()
    
    # Remove repeated input
    if response.lower() == message.lower():
        response = "Samahani, naweza kusaidia vipi?"

    return response