Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,233 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# app.py
|
2 |
+
|
3 |
+
import os
|
4 |
+
import openai
|
5 |
+
import gradio as gr
|
6 |
+
from langchain.chains import ConversationalRetrievalChain
|
7 |
+
from langchain.text_splitter import CharacterTextSplitter
|
8 |
+
from langchain.document_loaders import PyMuPDFLoader, PyPDFLoader
|
9 |
+
from langchain.vectorstores import Chroma
|
10 |
+
from langchain.embeddings.openai import OpenAIEmbeddings
|
11 |
+
from langchain.chat_models import ChatOpenAI
|
12 |
+
import shutil # 用於文件複製
|
13 |
+
|
14 |
+
# 獲取 OpenAI API 密鑰
|
15 |
+
api_key = os.getenv("OPENAI_API_KEY")
|
16 |
+
if not api_key:
|
17 |
+
raise ValueError("未能獲取 OpenAI_API_KEY。請在 Hugging Face Spaces 的 Secrets 中設置它。")
|
18 |
+
openai.api_key = api_key
|
19 |
+
print("OpenAI API 密鑰已設置。")
|
20 |
+
|
21 |
+
# 確保向量資料庫目錄存在且有寫入權限
|
22 |
+
VECTORDB_DIR = os.path.abspath("./data")
|
23 |
+
os.makedirs(VECTORDB_DIR, exist_ok=True)
|
24 |
+
os.chmod(VECTORDB_DIR, 0o755) # 設置適當的權限
|
25 |
+
print(f"VECTORDB_DIR set to: {VECTORDB_DIR}")
|
26 |
+
|
27 |
+
# 定義測試 PDF 加載器的函數
|
28 |
+
def test_pdf_loader(file_path, loader_type='PyMuPDFLoader'):
|
29 |
+
print(f"Testing PDF loader ({loader_type}) with file: {file_path}")
|
30 |
+
try:
|
31 |
+
if loader_type == 'PyMuPDFLoader':
|
32 |
+
loader = PyMuPDFLoader(file_path)
|
33 |
+
elif loader_type == 'PyPDFLoader':
|
34 |
+
loader = PyPDFLoader(file_path)
|
35 |
+
else:
|
36 |
+
print(f"Unknown loader type: {loader_type}")
|
37 |
+
return
|
38 |
+
loaded_docs = loader.load()
|
39 |
+
if loaded_docs:
|
40 |
+
print(f"Successfully loaded {file_path} with {len(loaded_docs)} documents.")
|
41 |
+
print(f"Document content (first 500 chars): {loaded_docs[0].page_content[:500]}")
|
42 |
+
else:
|
43 |
+
print(f"No documents loaded from {file_path}.")
|
44 |
+
except Exception as e:
|
45 |
+
print(f"Error loading {file_path} with {loader_type}: {e}")
|
46 |
+
|
47 |
+
# 定義載入和處理 PDF 文件的函數
|
48 |
+
def load_and_process_documents(file_paths, loader_type='PyMuPDFLoader'):
|
49 |
+
documents = []
|
50 |
+
print("開始載入上傳的 PDF 文件。")
|
51 |
+
|
52 |
+
for file_path in file_paths:
|
53 |
+
print(f"載入 PDF 文件: {file_path}")
|
54 |
+
if not os.path.exists(file_path):
|
55 |
+
print(f"文件不存在: {file_path}")
|
56 |
+
continue
|
57 |
+
try:
|
58 |
+
if loader_type == 'PyMuPDFLoader':
|
59 |
+
loader = PyMuPDFLoader(file_path)
|
60 |
+
elif loader_type == 'PyPDFLoader':
|
61 |
+
loader = PyPDFLoader(file_path)
|
62 |
+
else:
|
63 |
+
print(f"Unknown loader type: {loader_type}")
|
64 |
+
continue
|
65 |
+
loaded_docs = loader.load()
|
66 |
+
if loaded_docs:
|
67 |
+
print(f"載入 {file_path} 成功,包含 {len(loaded_docs)} 個文檔。")
|
68 |
+
# 打印第一個文檔的部分內容以確認
|
69 |
+
print(f"第一個文檔內容: {loaded_docs[0].page_content[:500]}")
|
70 |
+
documents.extend(loaded_docs)
|
71 |
+
else:
|
72 |
+
print(f"載入 {file_path} 但未找到任何文檔。")
|
73 |
+
except Exception as e:
|
74 |
+
print(f"載入 {file_path} 時出現錯誤: {e}")
|
75 |
+
|
76 |
+
if not documents:
|
77 |
+
raise ValueError("沒有找到任何 PDF 文件或 PDF 文件無法載入。")
|
78 |
+
else:
|
79 |
+
print(f"總共載入了 {len(documents)} 個文檔。")
|
80 |
+
|
81 |
+
# 分割長文本
|
82 |
+
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=50)
|
83 |
+
documents = text_splitter.split_documents(documents)
|
84 |
+
print(f"分割後的文檔數量: {len(documents)}")
|
85 |
+
|
86 |
+
if not documents:
|
87 |
+
raise ValueError("分割後的文檔列表為空。請檢查 PDF 文件內容。")
|
88 |
+
|
89 |
+
# 初始化向量資料庫
|
90 |
+
try:
|
91 |
+
embeddings = OpenAIEmbeddings(openai_api_key=api_key) # 直接傳遞 API 密鑰
|
92 |
+
print("初始化 OpenAIEmbeddings 成功。")
|
93 |
+
except Exception as e:
|
94 |
+
raise ValueError(f"初始化 OpenAIEmbeddings 時出現錯誤: {e}")
|
95 |
+
|
96 |
+
try:
|
97 |
+
vectordb = Chroma.from_documents(
|
98 |
+
documents,
|
99 |
+
embedding=embeddings,
|
100 |
+
persist_directory=VECTORDB_DIR
|
101 |
+
)
|
102 |
+
print("初始化 Chroma 向量資料庫成功。")
|
103 |
+
except Exception as e:
|
104 |
+
raise ValueError(f"初始化 Chroma 向量資料庫時出現錯誤: {e}")
|
105 |
+
|
106 |
+
return vectordb
|
107 |
+
|
108 |
+
# 定義聊天處理函數
|
109 |
+
def handle_query(user_message, chat_history, vectordb):
|
110 |
+
try:
|
111 |
+
if not user_message:
|
112 |
+
return chat_history
|
113 |
+
|
114 |
+
# 添加角色指令前綴
|
115 |
+
preface = """
|
116 |
+
指令: 以繁體中文回答問題,200字以內。你是一位專業心理學家與調酒師,專精於 MBTI 人格與經典調酒主題。
|
117 |
+
非相關問題,請回應:「目前僅支援 MBTI 分析與經典調酒主題。」。
|
118 |
+
"""
|
119 |
+
query = f"{preface} 查詢內容:{user_message}"
|
120 |
+
|
121 |
+
# 初始化 ConversationalRetrievalChain,並傳遞 openai_api_key
|
122 |
+
pdf_qa = ConversationalRetrievalChain.from_llm(
|
123 |
+
ChatOpenAI(temperature=0.7, model="gpt-4", openai_api_key=api_key),
|
124 |
+
retriever=vectordb.as_retriever(search_kwargs={'k': 6}),
|
125 |
+
return_source_documents=True
|
126 |
+
)
|
127 |
+
|
128 |
+
# 呼叫模型並處理查詢
|
129 |
+
result = pdf_qa.invoke({"question": query, "chat_history": chat_history})
|
130 |
+
|
131 |
+
# 檢查結果並更新聊天歷史
|
132 |
+
if "answer" in result:
|
133 |
+
chat_history = chat_history + [(user_message, result["answer"])]
|
134 |
+
else:
|
135 |
+
chat_history = chat_history + [(user_message, "抱歉,未能獲得有效回應。")]
|
136 |
+
return chat_history
|
137 |
+
|
138 |
+
except Exception as e:
|
139 |
+
return chat_history + [("系統", f"出現錯誤: {str(e)}")]
|
140 |
+
|
141 |
+
# 定義 Gradio 的處理函數
|
142 |
+
def process_files(files, state):
|
143 |
+
print("process_files called")
|
144 |
+
if files:
|
145 |
+
try:
|
146 |
+
print(f"Received {len(files)} files")
|
147 |
+
saved_file_paths = []
|
148 |
+
for file_path in files:
|
149 |
+
print(f"Processing file: {file_path}")
|
150 |
+
save_path = os.path.join(VECTORDB_DIR, os.path.basename(file_path))
|
151 |
+
# 複製文件到 VECTORDB_DIR
|
152 |
+
shutil.copy(file_path, save_path)
|
153 |
+
# 確認文件是否存在
|
154 |
+
if os.path.exists(save_path):
|
155 |
+
print(f"File successfully saved to: {save_path}")
|
156 |
+
else:
|
157 |
+
print(f"Failed to save file to: {save_path}")
|
158 |
+
saved_file_paths.append(save_path)
|
159 |
+
# 測試 PDF 加載器
|
160 |
+
test_pdf_loader(save_path, loader_type='PyMuPDFLoader')
|
161 |
+
# 列出 VECTORDB_DIR 中的所有文件
|
162 |
+
saved_files = os.listdir(VECTORDB_DIR)
|
163 |
+
print(f"Files in VECTORDB_DIR ({VECTORDB_DIR}): {saved_files}")
|
164 |
+
vectordb = load_and_process_documents(saved_file_paths, loader_type='PyMuPDFLoader')
|
165 |
+
state['vectordb'] = vectordb
|
166 |
+
return "PDF 文件已成功上傳並處理。您現在可以開始提問。", state
|
167 |
+
except Exception as e:
|
168 |
+
print(f"Error in process_files: {e}")
|
169 |
+
return f"處理文件時出現錯誤: {e}", state
|
170 |
+
else:
|
171 |
+
return "請上傳至少一個 PDF 文件。", state
|
172 |
+
|
173 |
+
def chat_interface(user_message, chat_history, state):
|
174 |
+
vectordb = state.get('vectordb', None)
|
175 |
+
if not vectordb:
|
176 |
+
return chat_history, state, "請先上傳 PDF 文件以進行處理。"
|
177 |
+
|
178 |
+
# 處理查詢
|
179 |
+
updated_history = handle_query(user_message, chat_history, vectordb)
|
180 |
+
return updated_history, state, ""
|
181 |
+
|
182 |
+
# 設計 Gradio 介面
|
183 |
+
with gr.Blocks() as demo:
|
184 |
+
gr.Markdown("<h1 style='text-align: center;'>MBTI 與經典調酒 AI 助理</h1>")
|
185 |
+
|
186 |
+
# 定義共享的 state
|
187 |
+
state = gr.State({"vectordb": None})
|
188 |
+
|
189 |
+
with gr.Tab("上傳 PDF 文件"):
|
190 |
+
with gr.Row():
|
191 |
+
with gr.Column(scale=1):
|
192 |
+
upload = gr.File(
|
193 |
+
file_count="multiple",
|
194 |
+
file_types=[".pdf"],
|
195 |
+
label="上傳 PDF 文件",
|
196 |
+
interactive=True,
|
197 |
+
type="filepath" # 保持為 "filepath"
|
198 |
+
)
|
199 |
+
upload_btn = gr.Button("上傳並處理")
|
200 |
+
upload_status = gr.Textbox(label="上傳狀態", interactive=False)
|
201 |
+
|
202 |
+
with gr.Tab("聊天機器人"):
|
203 |
+
chatbot = gr.Chatbot()
|
204 |
+
|
205 |
+
with gr.Row():
|
206 |
+
with gr.Column(scale=0.85):
|
207 |
+
txt = gr.Textbox(show_label=False, placeholder="請輸入您的問題...")
|
208 |
+
with gr.Column(scale=0.15, min_width=0):
|
209 |
+
submit_btn = gr.Button("提問")
|
210 |
+
|
211 |
+
# 綁定提問按鈕
|
212 |
+
submit_btn.click(
|
213 |
+
chat_interface,
|
214 |
+
inputs=[txt, chatbot, state],
|
215 |
+
outputs=[chatbot, state, txt]
|
216 |
+
)
|
217 |
+
|
218 |
+
# 綁定輸入框的提交事件
|
219 |
+
txt.submit(
|
220 |
+
chat_interface,
|
221 |
+
inputs=[txt, chatbot, state],
|
222 |
+
outputs=[chatbot, state, txt]
|
223 |
+
)
|
224 |
+
|
225 |
+
# 綁定上傳按鈕
|
226 |
+
upload_btn.click(
|
227 |
+
process_files,
|
228 |
+
inputs=[upload, state],
|
229 |
+
outputs=[upload_status, state]
|
230 |
+
)
|
231 |
+
|
232 |
+
# 啟動 Gradio 應用
|
233 |
+
demo.launch()
|