app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("google/mt5-small")
model = AutoModelForSeq2SeqLM.from_pretrained("./checkpoint-15000/")


def text_processing(text):
    inputs = [text]

    # Tokenize and prepare the inputs for model
    input_ids = tokenizer(inputs, return_tensors="pt", max_length=512, truncation=True, padding="max_length").input_ids
    attention_mask = tokenizer(inputs, return_tensors="pt", max_length=512, truncation=True, padding="max_length").attention_mask

    # Generate prediction
    output = model.generate(input_ids=input_ids, attention_mask=attention_mask, max_new_tokens=512)

    # Decode the prediction
    decoded_output = [tokenizer.decode(ids, skip_special_tokens=True) for ids in output]

    return decoded_output[0]

iface = gr.Interface(fn = text_processing, inputs='text', outputs=['text'], title='Punctuation Mark Prediction', description='本模型主要用于语言识别模型输出的后处理。\n输入无符号句子，需要打标点处用空格隔开，返回带标点句子。\n仅支持中文，因为训练数据中只有中文。')

iface.launch(inline=False)