simple_chat.py

import warnings
warnings.simplefilter(action="ignore")

from transformers.utils import logging
logging.set_verbosity_error()

import torch
import intel_extension_for_pytorch as ipex
from transformers import AutoModelForCausalLM, AutoTokenizer


def chat_with_model(device="xpu"):
    model_path = "microsoft/DialoGPT-small"
    tokenizer_path = "microsoft/DialoGPT-small"
    model_tokenizer = AutoTokenizer.from_pretrained(tokenizer_path, padding_side="left")
    model = AutoModelForCausalLM.from_pretrained(model_path).to(torch.bfloat16)
    if device == "xpu":
        device = torch.device("xpu")
    else:
        device = torch.device("cpu")
    model.to(device)
    print("Chatbot Ready! Start chatting! (Type 'quit' to exit)")
    while True:
        user_input = input("You: ")
        if user_input.lower() == "quit":
            break
        if not user_input.strip():
            print("Please enter a valid message.")
            continue
        new_input_ids = model_tokenizer.encode(
            user_input + model_tokenizer.eos_token, return_tensors="pt"
        ).to(device)
        response_ids = model.generate(
            new_input_ids,
            max_length=150,
            pad_token_id=model_tokenizer.eos_token_id,
            do_sample=True,
            temperature=0.7,
            top_p=0.95,
        )
        response = model_tokenizer.decode(
            response_ids[:, new_input_ids.shape[-1] :][0], skip_special_tokens=True
        )
        print("Chatbot: ", response)


if __name__ == "__main__":
    chat_with_model()