from transformers import AutoTokenizer, AutoModelForCausalLM

# Load a small, fast model
model_name = "distilgpt2"  # smaller than full GPT-2, uses less memory
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

conversation = "You are a kind AI assistant. Stay on topic.\n"
print("\nType your messages below. Type 'quit' to exit.\n")

while True:
    try:
        user_input = input("You: ")
    except KeyboardInterrupt:
        print("\nEnding chat. Bye!")
        break

    if user_input.lower() in ["quit", "exit"]:
        print("Ending chat. Bye!")
        break

    conversation += f"User: {user_input}\nAI:"

    # Tokenize efficiently and limit input size to avoid memory issues
    inputs = tokenizer(conversation, return_tensors="pt", truncation=True, max_length=512)

    # Generate a short response to save memory and speed
    output = model.generate(
        **inputs,
        max_new_tokens=40,       # shorter responses = faster
        temperature=0.7,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )

    # Decode and extract AI response
    ai_response = tokenizer.decode(output[0], skip_special_tokens=True).split("AI:")[-1].strip()
    print(f"Baby AI: {ai_response}")

    # Add AI response to conversation for context
    conversation += f"{ai_response}\n"