app.py

import gradio as gr
from datasets import ClassLabel
from datasets import load_dataset
import random
import pandas as pd
from huggingface_hub import login

def remove_space(example):
    '''
    移除資料集當中「犯罪事實」欄 (Corpus-Delicti) 當中作為斷詞字元的空白字元，以及每句開頭的「ㄧ、」。
    並且在文章的開頭跟結尾加入 bos_token = '<s>', eos_token = '</s>'
    '''
    return {'Corpus-Delicti': example['Corpus-Delicti'].replace(" ", "").split('一、')[1]} 

def download_file(content, filename):
#    print(filename)
#    print(content)
    with open(filename, "w", encoding="utf-8") as f:
        f.write(content)

def random_elements(dataset, num_examples=5):
    assert num_examples <= len(dataset), "Can't pick more elements than there are in the dataset."
    picks = []
    for _ in range(num_examples):
        pick = random.randint(0, len(dataset)-1)
        while pick in picks:
            pick = random.randint(0, len(dataset)-1)
        picks.append(pick)
    
    df = pd.DataFrame(dataset[picks])
    for column, typ in dataset.features.items():
        if isinstance(typ, ClassLabel):
            df[column] = df[column].transform(lambda i: typ.names[i])
    return df

def random_next(num_examples=5):
    random_selected = random_elements(dataset["train"], num_examples=num_examples)
    court_name = random_selected['Court'][0]
    case_no = random_selected['CaseNo'][0]
    crime_descrip = random_selected['Corpus-Delicti'][0]
    title = court_name + "_" + case_no
    data_tuple = (court_name, case_no, crime_descrip, title)
    return data_tuple

def gen_template(crime_descrip, element, tag):
    INTRO_BLURB = "The following is a description of the crime in the verdict. Write a response for the element of crime and its tag that appropriately completes the request."
    DESCRIPT_KEY = "### Description:"
    ELEMENT_KEY = "### Element:"
    TAG_KEY = "### Tag:"
    END_KEY = "### End"
#    assert tag == None, "未選取構成要件要素標籤"
    try:
        tag_name = tag.split(",")[1].strip(")").strip().strip("'")
    except IndexError: # 防呆用的。如果什麼資料都沒填就按下按鈕，就會觸發以下程式碼，並傳回空樣板。
        # 改為調適 Alpaca 格式的資料
        blurb = f"{INTRO_BLURB}\n"
        descript = f"{DESCRIPT_KEY}\n{crime_descrip}\n"
        element = f"{ELEMENT_KEY}\n{element}\n" if element else f"{ELEMENT_KEY}\n<未填寫構成要件要素>\n"
        tag = f"{TAG_KEY}\n{tag_name}\n" if tag else f"{TAG_KEY}\n<未選取構成要件要素標籤>\n"
        end = f"{END_KEY}"
        template = blurb + '\n' + descript + '\n' + element + '\n' + tag + '\n' + end
        return template
    blurb = f"{INTRO_BLURB}\n"
    # 改為調適 Alpaca 格式的資料
    descript = f"{DESCRIPT_KEY}\n{crime_descrip}\n"
    element = f"{ELEMENT_KEY}\n{element}\n" if element else f"{ELEMENT_KEY}\n<未填寫構成要件要素>\n"
    tag = f"{TAG_KEY}\n{tag_name}\n" if tag else f"{TAG_KEY}\n<未選取構成要件要素標籤>\n"
    end = f"{END_KEY}"
    template = blurb + '\n' + descript + '\n' + element + '\n' + tag + '\n' + end
    return template

#random_selected = random_elements(dataset["train"])
random_selected = random_next()
court_name = random_selected[0]
case_no = random_selected[1]
crime_descrip = random_selected[2]
title = random_selected[3]

with gr.Blocks() as demo:
    gr.Markdown(
    """
    <h1 style="text-align: center;">Legal Document Annotation</h1>
    """)
    with gr.Row():
        with gr.Column(): # 犯罪事實段
#            court_name = random_selected[0]
#            case_no = random_selected[1]
#            crime_descrip = random_selected[2]
            with gr.Row(): # 抬頭段
#                courtName = gr.Label(label='法院名稱', value=court_name)
#                caseNo = gr.Label(label='案號', value=case_no)
                title = gr.components.Textbox(label='案號',value=title)
            prompt = gr.components.Textbox(lines=5, label='犯罪事實',value=crime_descrip)
            with gr.Row():
                with gr.Column():
                    btn = gr.Button("隨機選擇")
#            gr.Examples(examples, inputs=[prompt])
        with gr.Column():
            with gr.Row():
                element = gr.components.Textbox(lines=2, label="構成要件要素")
#                tag = gr.components.Textbox(label="標籤")
            tag = gr.Dropdown(
                              choices = [("被告(犯罪主體)","<LEO_SOC>"), ("主觀犯意", "<LEO_SLE>"), ("不法行為","<LEO_ACT>"), ("因果關係","<LEO_CAU>"),
                                         ("被害人/告訴人","<LEO_VIC>"), ("危害結果","<LEO_ROH>"), ("未遂","<LEO_ATP>"), ("既遂","<LEO_ACC>"),
                                         ("中止","<LEO_ABA>"), ("預備","<LEO_PRP>")], 
                              label="標籤", info="構成要件要素的標籤")
            with gr.Row():
                with gr.Column():
                    btn2 = gr.Button("產生標註語料內容")
            result = gr.components.Textbox(lines=5, label="語料內容", show_copy_button=True)
#            btn3 = gr.Button("下載")
        btn.click(random_next, inputs=[], outputs=[courtName, caseNo, prompt, title])
        btn2.click(gen_template, inputs=[prompt, element, tag], outputs=[result])
#        btn3.click(download_file, inputs=[result, title], outputs=[])

if __name__ == "__main__":
    # 下載判決書資料集
    login(token = os.environ['HUB_TOKEN'])
    use_auth_token=os.environ['HUB_TOKEN'] # 下載判決書資料集所需要的 token。
    dataset = load_dataset("jslin09/Fraud_Case_Verdicts", use_auth_token=use_auth_token, revision="main")
    dataset = dataset.map(remove_space)
    demo.launch(share=True) # 在遠端啟動時，需要 share=True 。