import torch
import gradio as gr
from transformers import XGLMTokenizer, XGLMForCausalLM

tokenizer = XGLMTokenizer.from_pretrained("facebook/xglm-1.7B")
model = XGLMForCausalLM.from_pretrained("facebook/xglm-1.7B", low_cpu_mem_usage=False)

def generate(text):
  inputs = tokenizer(text, return_tensors="pt")
  outputs = model.generate(**inputs, labels=inputs["input_ids"], 
  do_sample=True, max_length=2048)
  return tokenizer.decode(outputs.tolist()[0])

gr.Interface(fn=generate, inputs="text", outputs="text").launch()