Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,34 +1,84 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import pipeline
|
|
|
|
|
|
|
| 3 |
|
| 4 |
# 1. 載入 SQuAD v2.0 預訓練模型
|
|
|
|
| 5 |
qa_model = pipeline("question-answering", model="deepset/roberta-base-squad2")
|
| 6 |
|
| 7 |
-
# 2. 定義
|
| 8 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
if not context or not question:
|
| 10 |
-
return "請
|
| 11 |
|
| 12 |
-
# 執行問答
|
|
|
|
| 13 |
result = qa_model(question=question, context=context)
|
| 14 |
|
| 15 |
-
#
|
| 16 |
if result['score'] < 0.05:
|
| 17 |
-
return "抱歉,在文件中找不到相關答案。"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
|
| 19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
-
#
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
description="根據提供的文本回答問題。"
|
| 31 |
-
)
|
| 32 |
|
| 33 |
if __name__ == "__main__":
|
| 34 |
demo.launch()
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import pipeline
|
| 3 |
+
import pdfplumber
|
| 4 |
+
import docx
|
| 5 |
|
| 6 |
# 1. 載入 SQuAD v2.0 預訓練模型
|
| 7 |
+
# 使用 deepset/roberta-base-squad2,它是針對 v2.0 優化的標準模型
|
| 8 |
qa_model = pipeline("question-answering", model="deepset/roberta-base-squad2")
|
| 9 |
|
| 10 |
+
# 2. 定義文件讀取函式
|
| 11 |
+
def extract_text(file):
|
| 12 |
+
if file is None:
|
| 13 |
+
return ""
|
| 14 |
+
|
| 15 |
+
file_path = file.name
|
| 16 |
+
text = ""
|
| 17 |
+
|
| 18 |
+
# 處理 PDF
|
| 19 |
+
if file_path.endswith('.pdf'):
|
| 20 |
+
with pdfplumber.open(file_path) as pdf:
|
| 21 |
+
for page in pdf.pages:
|
| 22 |
+
text += page.extract_text() + "\n"
|
| 23 |
+
|
| 24 |
+
# 處理 Word (.docx)
|
| 25 |
+
elif file_path.endswith('.docx'):
|
| 26 |
+
doc = docx.Document(file_path)
|
| 27 |
+
for para in doc.paragraphs:
|
| 28 |
+
text += para.text + "\n"
|
| 29 |
+
|
| 30 |
+
# 處理純文字 (.txt)
|
| 31 |
+
elif file_path.endswith('.txt'):
|
| 32 |
+
with open(file_path, 'r', encoding='utf-8') as f:
|
| 33 |
+
text = f.read()
|
| 34 |
+
|
| 35 |
+
return text
|
| 36 |
+
|
| 37 |
+
# 3. 定義主預測邏輯
|
| 38 |
+
def predict(file, manual_context, question):
|
| 39 |
+
# 優先從上傳的文件提取內容,若無則使用手動輸入的內容
|
| 40 |
+
if file is not None:
|
| 41 |
+
context = extract_text(file)
|
| 42 |
+
else:
|
| 43 |
+
context = manual_context
|
| 44 |
+
|
| 45 |
if not context or not question:
|
| 46 |
+
return "請先提供文件內容(上傳或貼上文字)並輸入提問。"
|
| 47 |
|
| 48 |
+
# 執行問答推理
|
| 49 |
+
# 加入 handle_impossible_answer=True 處理 SQuAD v2.0 特性
|
| 50 |
result = qa_model(question=question, context=context)
|
| 51 |
|
| 52 |
+
# 信心門檻判斷
|
| 53 |
if result['score'] < 0.05:
|
| 54 |
+
return "抱歉,在文件內容中找不到相關答案(模型信心程度較低)。"
|
| 55 |
+
|
| 56 |
+
return f"回答:{result['answer']}\n(信心分數: {round(result['score'], 4)})"
|
| 57 |
+
|
| 58 |
+
# 4. 建立 Gradio 網頁介面
|
| 59 |
+
with gr.Blocks(title="Case Study: AI Document QA") as demo:
|
| 60 |
+
gr.Markdown("# 📑 Case Study: 智慧文件問答系統")
|
| 61 |
+
gr.Markdown("利用語言模型進行文件自動化讀取與問答。")
|
| 62 |
|
| 63 |
+
with gr.Row():
|
| 64 |
+
with gr.Column():
|
| 65 |
+
file_input = gr.File(label="1. 上傳文件 (PDF, Word, TXT)")
|
| 66 |
+
text_input = gr.Textbox(lines=8, label="或是在此貼上文件內容", placeholder="若已上傳文件則無需填寫此處...")
|
| 67 |
+
question_input = gr.Textbox(lines=2, label="2. 輸入您的問題", placeholder="例如:這份文件的主要結論是什麼?")
|
| 68 |
+
submit_btn = gr.Button("開始分析", variant="primary")
|
| 69 |
+
|
| 70 |
+
with gr.Column():
|
| 71 |
+
answer_output = gr.Textbox(label="模型回答結果", lines=10)
|
| 72 |
|
| 73 |
+
# 綁定按鈕功能
|
| 74 |
+
submit_btn.click(
|
| 75 |
+
fn=predict,
|
| 76 |
+
inputs=[file_input, text_input, question_input],
|
| 77 |
+
outputs=answer_output
|
| 78 |
+
)
|
| 79 |
+
|
| 80 |
+
gr.Markdown("---")
|
| 81 |
+
gr.Markdown("💡 **提示:** 針對 SQuAD v2.0 資料集訓練的模型具備判斷『問題是否可回答』的能力。")
|
|
|
|
|
|
|
| 82 |
|
| 83 |
if __name__ == "__main__":
|
| 84 |
demo.launch()
|