app.py

import gradio as gr
from transformers import PretrainedTokenizerFast, BartForCondtionalGeneration

model_name = 'ainize/kobart-news'
tokenizer = PreTrainedTokenizerFast.from_pretrained(model_name)
model = BartForConditionalGeneration.from_pretrained(model_name)

def summ(txt):
  input_ids = tokenizer.encode(txt, return_tensors="pt")
  summary_text_ids = model.generate(
    input_ids=input_ids,
    bos_token_id=model.config.bos_token_id, # BOS는 Beginning of Sentence
    eos_token_id=model.config.eos_token_id, # EOS는 End of Sentence
    length_penalty=2.0, # 요약을 얼마나 짧게 할지
    max_length=142, # 최대 142 토큰으로 요약
    min_length=56,  # 최소 56 토큰보다는 더 나옴
    num_beams=4    # beam search
  )
  return tokenizer.decode(summary_text_ids[0], skip_special_tokens=True)

interface = gr.Interface(summ,
                        [gr.Textbox(label="original text")], # gradio를 이용해서 gui
                        [gr.Textbox(label="summary")])

interface.launch() # share가 True일 경우, link가 하나 나옴