app.py

"""
baseline_interactive.py
"""
import gradio as gr
from transformers import MBartForConditionalGeneration, MBartTokenizer
from transformers import pipeline

model_name = "momo/rsp-sum"
model = MBartForConditionalGeneration.from_pretrained(model_name)
tokenizer = MBartTokenizer.from_pretrained(model_name, src_lang="ko_KR", tgt_lang="ko_KR")

summarizer = pipeline("summarization", model=model, tokenizer=tokenizer)

def summarization(News):
    result = summarizer(News, min_length=50, max_length=150)
    return result[0]["summary_text"]

if __name__ == '__main__':
    app = gr.Interface(
        fn=summarization,
        inputs=gr.inputs.Textbox(lines=10, label="News"),
        outputs=gr.outputs.Textbox(label="Summary"), 
        title="한국어 뉴스 요약 생성기",
        description="Korean News Summary Generator"
        )
    app.launch()

# with torch.no_grad():
#     while True:
#         t = input("\nDocument: ")
#         tokens = tokenizer(
#             t,
#             return_tensors="pt",
#             truncation=True,
#             padding=True,
#             max_length=600
#         )

#         input_ids = tokens.input_ids.cuda()
#         attention_mask = tokens.attention_mask.cuda()

#         sample_output = model.generate(
#             input_ids, 
#             max_length=150, 
#             num_beams=5, 
#             early_stopping=True,
#             no_repeat_ngram_size=8,
#     )
#         # print("token:" + str(input_ids.detach().cpu()))
#         # print("token:" + tokenizer.convert_ids_to_tokens(str(input_ids.detach().cpu())))
#         print("Summary: " + tokenizer.decode(sample_output[0], skip_special_tokens=True))