from transformers import AutoModelForCausalLM, AutoTokenizer
import gradio as gr
import torch

# DialoGPT Modell und Tokenizer laden
tokenizer = AutoTokenizer.from_pretrained("microsoft/DialoGPT-medium")
model = AutoModelForCausalLM.from_pretrained("microsoft/DialoGPT-medium")

def chat_with_kiki_gpt(user_input):
    # Benutzereingabe kodieren und Modellantwort generieren
    input_ids = tokenizer.encode(user_input + tokenizer.eos_token, return_tensors='pt')
    chat_history_ids = model.generate(input_ids, max_length=1000, pad_token_id=tokenizer.eos_token_id)
    chat_output = tokenizer.decode(chat_history_ids[:, input_ids.shape[-1]:][0], skip_special_tokens=True)
    
    return f"KIKI-GPT: {chat_output}"

# Gradio-Benutzeroberfläche
interface = gr.Interface(
    fn=chat_with_kiki_gpt,
    inputs=gr.inputs.Textbox(lines=5, placeholder="Type your message to KIKI-GPT here..."),
    outputs=gr.outputs.Textbox(),
    title="KIKI-GPT",
    description="Welcome to KIKI-GPT - a project on Hugging Face Spaces using Microsoft's DialoGPT. One of the fastest and best performing models for robotics! Created by Keyvan Hardani. For inquiries, contact: hello@keyvan.ai.",
    live=True
)

interface.launch()