app.py

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

tokenizer = AutoTokenizer.from_pretrained("microsoft/DialoGPT-large")
model = AutoModelForCausalLM.from_pretrained("microsoft/DialoGPT-large")

def predict(input, history=[]):
  new_user_input_ids = tokenizer.encode(input + tokenizer.eos_token, return_tensors='pt')
  bot_input_ids = torch.cat([torch.LongTensor(history), new_user_input_ids], dim=-1)
  history = model.generate(bot_input_ids, max_length=500, pad_token_id=tokenizer.eos_token_id).tolist()
  response = tokenizer.decode(history[0]).replace("<|endoftext|>", "\n")
  
  return response, history

gr.Interface(fn=predict, title="ineuronbot", inputs=[gr.inputs.Textbox(placeholder="Write a text"), "state"], 
             outputs=[gr.outputs.Textbox(label="Output"), "state"]).launch()