Add database.

README.md

@@ -1,5 +1,5 @@
 ---
-title: LightZero RAG
+title: ZeroPal
 emoji: 📖
 colorFrom: yellow
 colorTo: blue
@@ -58,7 +58,7 @@ QUESTION_LANG='cn' # The language of the question, currently available option is
 
 if __name__ == "__main__":
     # Assuming documents are already present locally
-    file_path = './documents/LightZero_README.zh.md'
+    file_path = './documents/LightZero_README_zh.md'
     # Load and split document
     chunks = load_and_split_document(file_path)
     # Create vector store

README_zh.md

@@ -45,7 +45,7 @@ QUESTION_LANG='cn' # 问题语言，目前可选值为 'cn'
 
 if __name__ == "__main__":
     # 假设文档已存在于本地
-    file_path = './documents/LightZero_README.zh.md'
+    file_path = './documents/LightZero_README_zh.md'
     # 加载和分割文档
     chunks = load_and_split_document(file_path)
     # 创建向量存储

analyze_conversation_history.py

@@ -0,0 +1,68 @@
+import sqlite3
+
+
+def analyze_conversation_history():
+    """
+    分析对话历史数据库中的数据
+    """
+    # 连接到SQLite数据库
+    conn = sqlite3.connect('database/conversation_history.db')
+    c = conn.cursor()
+
+    # 获取总的对话记录数
+    c.execute("SELECT COUNT(*) FROM history")
+    total_records = c.fetchone()[0]
+    print(f"总对话记录数: {total_records}")
+
+    # 获取不同用户的对话记录数
+    c.execute("SELECT user_id, COUNT(*) as count FROM history GROUP BY user_id")
+    user_records = c.fetchall()
+    print("每个用户的对话记录数:")
+    for user_id, count in user_records:
+        print(f"用户 {user_id}: {count} 条记录")
+
+    # 获取平均对话轮数
+    c.execute("SELECT AVG(cnt) FROM (SELECT user_id, COUNT(*) as cnt FROM history GROUP BY user_id)")
+    avg_turns = c.fetchone()[0]
+    print(f"平均对话轮数: {avg_turns}")
+
+    # 获取最长的用户输入和助手输出
+    c.execute("SELECT MAX(LENGTH(user_input)) FROM history")
+    max_user_input_length = c.fetchone()[0]
+    print(f"最长的用户输入: {max_user_input_length} 个字符")
+
+    c.execute("SELECT MAX(LENGTH(assistant_output)) FROM history")
+    max_assistant_output_length = c.fetchone()[0]
+    print(f"最长的助手输出: {max_assistant_output_length} 个字符")
+
+    # 关闭游标
+    c.close()
+    # 关闭数据库连接
+    conn.close()
+
+
+def clear_context():
+    """
+    清除对话历史
+    """
+    # 连接到SQLite数据库
+    conn = sqlite3.connect('conversation_history.db')
+    c = conn.cursor()
+    c.execute("DELETE FROM history")
+    conn.commit()
+    return "", "", ""
+
+
+def get_history():
+    """
+    获取对话历史记录
+    """
+    # 连接到SQLite数据库
+    conn = sqlite3.connect('conversation_history.db')
+    c = conn.cursor()
+    c.execute("SELECT user_input, assistant_output FROM history")
+    rows = c.fetchall()
+    history = ""
+    for row in rows:
+        history += f"User: {row[0]}\nAssistant: {row[1]}\n\n"
+    return history

app.py

@@ -1,9 +1,12 @@
 import os
+import sqlite3
+import threading
 
 import gradio as gr
 from dotenv import load_dotenv
 from langchain.document_loaders import TextLoader
 
+from RAG.analyze_conversation_history import analyze_conversation_history
 from rag_demo import load_and_split_document, create_vector_store, setup_rag_chain, execute_query
 
 # 环境设置
@@ -12,122 +15,200 @@ QUESTION_LANG = os.getenv("QUESTION_LANG")  # 从环境变量获取 QUESTION_LAN
 assert QUESTION_LANG in ['cn', 'en'], QUESTION_LANG
 
 if QUESTION_LANG == "cn":
-    title = "LightZero RAG Demo"
+    title = "ZeroPal"
     title_markdown = """
     <div align="center">
         <img src="https://raw.githubusercontent.com/puyuan1996/RAG/main/assets/banner.svg" width="80%" height="20%" alt="Banner Image">
     </div>
-    <h2 style="text-align: center; color: black;"><a href="https://github.com/puyuan1996/RAG"> LightZero RAG Demo</a></h2>
-    <h4 align="center"> 📢说明：请您在下面的"问题（Q）"框中输入任何关于 LightZero 的问题，然后点击"提交"按钮。右侧"回答（A）"框中会显示 RAG 模型给出的回答。在 QA 栏的下方会给出参考文档（其中检索得到的相关文段会用黄色高亮显示）。</h4>
-    <h4 align="center"> 如果你喜欢这个项目，请给我们在 GitHub 点个 star ✨ 。我们将会持续保持更新。  </h4>
-    <strong><h5 align="center">注意：算法模型的输出可能包含一定的随机性。相关结果不代表任何开发者和相关 AI 服务的态度和意见。本项目开发者不对生成结果作任何保证，仅供参考。<h5></strong>
+    
+    📢 **操作说明**：请在下方的“问题”框中输入关于 LightZero 的问题，并点击“提交”按钮。右侧的“回答”框将展示 RAG 模型提供的答案。
+    您可以在问答框下方查看当前“对话历史”，点击“清除上下文”按钮可清空历史记录。在“对话历史”框下方，您将找到相关参考文档，其中相关文段将以黄色高亮显示。
+    如果您喜欢这个项目，请在 GitHub [LightZero RAG Demo](https://github.com/puyuan1996/RAG) 上给我们点赞！✨ 您的支持是我们持续更新的动力。
+    
+    <div align="center">
+        <strong>注意：算法模型输出可能包含一定的随机性。结果不代表开发者和相关 AI 服务的态度和意见。本项目开发者不对结果作出任何保证，仅供参考之用。使用该服务即代表同意后文所述的使用条款。</strong>
+    </div>
     """
     tos_markdown = """
     ### 使用条款
-    玩家使用本服务须同意以下条款：
-    该服务是一项探索性研究预览版，仅供非商业用途。它仅提供有限的安全措施，并可能生成令人反感的内容。不得将其用于任何非法、有害、暴力、种族主义等目的。
-    如果您的游玩体验有不佳之处，请发送邮件至 [email protected] ！ 我们将删除相关信息，并不断改进这个项目。
-    为了获得最佳体验，请使用台式电脑，因为移动设备可能会影响可视化效果。
-    **版权所有 2024 OpenDILab。**
+    
+    使用本服务的玩家需同意以下条款：
+    
+    - 本服务为探索性研究的预览版，仅供非商业用途。
+    - 服务不得用于任何非法、有害、暴力、种族主义或其他令人反感的目的。
+    - 服务提供有限的安全措施，并可能生成令人反感的内容。
+    - 如果您对服务体验不满，请通过 [email protected] 与我们联系！我们承诺修复问题并不断改进项目。
+    - 为了获得最佳体验，请使用台式电脑，因为移动设备可能会影响视觉效果。
+    
+    **版权所有 © 2024 OpenDILab。保留所有权利。**
     """
 
-# 路径变量，方便之后的文件使用
-file_path = './documents/LightZero_README.zh.md'
+# 路径变量,方便之后的文件使用
+file_path = './documents/LightZero_README_zh.md'
 
 # 加载原始Markdown文档
 loader = TextLoader(file_path)
 orig_documents = loader.load()
 
 # 存储对话历史
-conversation_history = []
+conversation_history = {}
+
+# 创建线程局部数据对象
+threadLocal = threading.local()
+
+
+def get_db_connection():
+    """
+    返回当前线程的数据库连接
+    """
+    conn = getattr(threadLocal, 'conn', None)
+    if conn is None:
+        # 连接到SQLite数据库
+        conn = sqlite3.connect('database/conversation_history.db')
+        c = conn.cursor()
+        # Drop the existing 'history' table if it exists
+        # c.execute('DROP TABLE IF EXISTS history')
+        # 创建存储对话历史的表
+        c.execute('''CREATE TABLE IF NOT EXISTS history
+                     (id INTEGER PRIMARY KEY AUTOINCREMENT,
+                     user_id TEXT NOT NULL,
+                     user_input TEXT NOT NULL,
+                     assistant_output TEXT NOT NULL,
+                     timestamp DATETIME DEFAULT CURRENT_TIMESTAMP)''')
+        threadLocal.conn = conn
+    return conn
+
+
+def get_db_cursor():
+    """
+    返回当前线程的数据库游标
+    """
+    conn = get_db_connection()
+    c = getattr(threadLocal, 'cursor', None)
+    if c is None:
+        c = conn.cursor()
+        threadLocal.cursor = c
+    return c
+
+
+# 程序结束时清理数据库连接
+def close_db_connection():
+    conn = getattr(threadLocal, 'conn', None)
+    if conn is not None:
+        conn.close()
+        setattr(threadLocal, 'conn', None)
+
+    c = getattr(threadLocal, 'cursor', None)
+    if c is not None:
+        c.close()
+        setattr(threadLocal, 'cursor', None)
 
 
-def rag_answer(question, model_name, temperature, embedding_model, k):
+def rag_answer(question, temperature, k, user_id):
     """
     处理用户问题并返回答案和高亮显示的上下文
 
     :param question: 用户输入的问题
-    :param model_name: 使用的语言模型名称
     :param temperature: 生成答案时使用的温度参数
-    :param embedding_model: 使用的嵌入模型
     :param k: 检索到的文档块数量
+    :param user_id: 用户ID
     :return: 模型生成的答案和高亮显示上下文的Markdown文本
     """
     try:
         chunks = load_and_split_document(file_path, chunk_size=5000, chunk_overlap=500)
-        retriever = create_vector_store(chunks, model=embedding_model, k=k)
-        rag_chain = setup_rag_chain(model_name=model_name, temperature=temperature)
+        retriever = create_vector_store(chunks, model='OpenAI', k=k)
+        rag_chain = setup_rag_chain(model_name='kimi', temperature=temperature)
 
-        # 将问题添加到对话历史中
-        conversation_history.append(("User", question))
+        if user_id not in conversation_history:
+            conversation_history[user_id] = []
 
-        # 将对话历史转换为字符串
-        history_str = "\n".join([f"{role}: {text}" for role, text in conversation_history])
+        conversation_history[user_id].append((f"User[{user_id}]", question))
 
-        retrieved_documents, answer = execute_query(retriever, rag_chain, history_str, model_name=model_name,
+        history_str = "\n".join([f"{role}: {text}" for role, text in conversation_history[user_id]])
+
+        retrieved_documents, answer = execute_query(retriever, rag_chain, history_str, model_name='kimi',
                                                     temperature=temperature)
+
+        ############################
+        # 获取当前线程的数据库连接和游标
+        ############################
+        conn = get_db_connection()
+        c = get_db_cursor()
+
+        # 分析对话历史
+        # analyze_conversation_history()
+        # 获取总的对话记录数
+        c.execute("SELECT COUNT(*) FROM history")
+        total_records = c.fetchone()[0]
+        print(f"总对话记录数: {total_records}")
+
+        # 将问题和回答存储到数据库
+        c.execute("INSERT INTO history (user_id, user_input, assistant_output) VALUES (?, ?, ?)",
+                  (user_id, question, answer))
+        conn.commit()
+
         # 在文档中高亮显示上下文
         context = [retrieved_documents[i].page_content for i in range(len(retrieved_documents))]
         highlighted_document = orig_documents[0].page_content
         for i in range(len(context)):
             highlighted_document = highlighted_document.replace(context[i], f"<mark>{context[i]}</mark>")
 
-        # 将回答添加到对话历史中
-        conversation_history.append(("Assistant", answer))
+        conversation_history[user_id].append(("Assistant", answer))
+
+        full_history = "\n".join([f"{role}: {text}" for role, text in conversation_history[user_id]])
     except Exception as e:
         print(f"An error occurred: {e}")
-        return "处理您的问题时出现错误，请稍后再试。", ""
-    return answer, highlighted_document
+        return "处理您的问题时出现错误,请稍后再试。", "", ""
+    finally:
+        # 不再在这里关闭游标和连接
+        pass
 
+    return answer, highlighted_document, full_history
 
-def clear_context():
+
+def clear_context(user_id):
     """
     清除对话历史
     """
-    global conversation_history
-    conversation_history = []
-    return "", ""
+    if user_id in conversation_history:
+        conversation_history[user_id] = []
+    return "", "", ""
 
 
 if __name__ == "__main__":
-    with gr.Blocks(title=title, theme='ParityError/Interstellar') as rag_demo:
+    with gr.Blocks(title=title, theme='ParityError/Interstellar') as zero_pal:
         gr.Markdown(title_markdown)
 
         with gr.Row():
             with gr.Column():
+                user_id = gr.Textbox(
+                    placeholder="请输入您的真实姓名或昵称作为用户ID",
+                    label="用户ID")
                 inputs = gr.Textbox(
-                    placeholder="请您输入任何关于 LightZero 的问题。",
-                    label="问题 (Q)")
-                model_name = gr.Dropdown(
-                    choices=['kimi', 'abab6-chat', 'glm-4', 'gpt-3.5-turbo', 'gpt-4', 'gpt-4-turbo', 'azure_gpt-4', 'azure_gpt-35-turbo-16k', 'azure_gpt-35-turbo'],
-                    # value='azure_gpt-4',
-                    value='kimi',
-                    label="选择语言模型")
+                    placeholder="请您在这里输入任何关于 LightZero 的问题。",
+                    label="问题")
                 temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.01, step=0.01, label="温度参数")
-                embedding_model = gr.Dropdown(
-                    choices=['HuggingFace', 'TensorflowHub', 'OpenAI'],
-                    value='OpenAI',
-                    label="选择嵌入模型")
                 k = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="检索到的文档块数量")
                 with gr.Row():
                     gr_submit = gr.Button('提交')
                     gr_clear = gr.Button('清除上下文')
 
-            outputs_answer = gr.Textbox(placeholder="当你点击提交按钮后，这里会显示 RAG 模型给出的回答。",
-                                        label="回答 (A)")
+            outputs_answer = gr.Textbox(placeholder="当你点击提交按钮后,这里会显示 RAG 模型给出的回答。",
+                                        label="回答")
+        outputs_history = gr.Textbox(label="对话历史")
         with gr.Row():
-            outputs_context = gr.Markdown(label="参考的文档，检索得到的 context 用高亮显示 (C)")
-
-        gr.Markdown(tos_markdown)
-
+            outputs_context = gr.Markdown(label="参考的文档(检索得到的相关文段用高亮显示)")
+        gr_clear.click(clear_context, inputs=user_id, outputs=[outputs_context, outputs_history])
         gr_submit.click(
             rag_answer,
-            inputs=[inputs, model_name, temperature, embedding_model, k],
-            outputs=[outputs_answer, outputs_context],
+            inputs=[inputs, temperature, k, user_id],
+            outputs=[outputs_answer, outputs_context, outputs_history],
         )
-        gr_clear.click(clear_context, outputs=[outputs_answer, outputs_context])
+        gr.Markdown(tos_markdown)
 
     concurrency = int(os.environ.get('CONCURRENCY', os.cpu_count()))
     favicon_path = os.path.join(os.path.dirname(__file__), 'assets', 'avatar.png')
-    rag_demo.queue().launch(max_threads=concurrency, favicon_path=favicon_path, share=True)
-    
+    zero_pal.queue().launch(max_threads=concurrency, favicon_path=favicon_path, share=True)
+
+    # 在合适的地方，例如程序退出时，调用close_db_connection函数
+    close_db_connection()

app_mqa.py

@@ -1,9 +1,7 @@
 import os
-
 import gradio as gr
 from dotenv import load_dotenv
 from langchain.document_loaders import TextLoader
-
 from rag_demo import load_and_split_document, create_vector_store, setup_rag_chain, execute_query
 
 # 环境设置
@@ -12,27 +10,36 @@ QUESTION_LANG = os.getenv("QUESTION_LANG")  # 从环境变量获取 QUESTION_LAN
 assert QUESTION_LANG in ['cn', 'en'], QUESTION_LANG
 
 if QUESTION_LANG == "cn":
-    title = "LightZero RAG Demo"
+    title = "ZeroPal"
     title_markdown = """
     <div align="center">
         <img src="https://raw.githubusercontent.com/puyuan1996/RAG/main/assets/banner.svg" width="80%" height="20%" alt="Banner Image">
     </div>
-    <h2 style="text-align: center; color: black;"><a href="https://github.com/puyuan1996/RAG"> LightZero RAG Demo</a></h2>
-    <h4 align="center"> 📢说明：请您在下面的"问题（Q）"框中输入任何关于 LightZero 的问题，然后点击"提交"按钮。右侧"回答（A）"框中会显示 RAG 模型给出的回答。在 QA 栏的下方会给出参考文档（其中检索得到的相关文段会用黄色高亮显示）。</h4>
-    <h4 align="center"> 如果你喜欢这个项目，请给我们在 GitHub 点个 star ✨ 。我们将会持续保持更新。  </h4>
-    <strong><h5 align="center">注意：算法模型的输出可能包含一定的随机性。相关结果不代表任何开发者和相关 AI 服务的态度和意见。本项目开发者不对生成结果作任何保证，仅供参考。<h5></strong>
+    
+    📢 **操作说明**：请在下方的“问题”框中输入关于 LightZero 的问题，并点击“提交”按钮。右侧的“回答”框将展示 RAG 模型提供的答案。
+    您可以在问答框下方查看当前“对话历史”，点击“清除上下文”按钮可清空历史记录。在“对话历史”框下方，您将找到相关参考文档，其中相关文段将以黄色高亮显示。
+    如果您喜欢这个项目，请在 GitHub [LightZero RAG Demo](https://github.com/puyuan1996/RAG) 上给我们点赞！✨ 您的支持是我们持续更新的动力。
+    
+    <div align="center">
+        <strong>注意：算法模型输出可能包含一定的随机性。结果不代表开发者和相关 AI 服务的态度和意见。本项目开发者不对结果作出任何保证，仅供参考之用。使用该服务即代表同意后文所述的使用条款。</strong>
+    </div>
     """
     tos_markdown = """
     ### 使用条款
-    玩家使用本服务须同意以下条款：
-    该服务是一项探索性研究预览版，仅供非商业用途。它仅提供有限的安全措施，并可能生成令人反感的内容。不得将其用于任何非法、有害、暴力、种族主义等目的。
-    如果您的游玩体验有不佳之处，请发送邮件至 [email protected] ！ 我们将删除相关信息，并不断改进这个项目。
-    为了获得最佳体验，请使用台式电脑，因为移动设备可能会影响可视化效果。
-    **版权所有 2024 OpenDILab。**
+    
+    使用本服务的玩家需同意以下条款：
+    
+    - 本服务为探索性研究的预览版，仅供非商业用途。
+    - 服务不得用于任何非法、有害、暴力、种族主义或其他令人反感的目的。
+    - 服务提供有限的安全措施，并可能生成令人反感的内容。
+    - 如果您对服务体验不满，请通过 [email protected] 与我们联系！我们承诺修复问题并不断改进项目。
+    - 为了获得最佳体验，请使用台式电脑，因为移动设备可能会影响视觉效果。
+    
+    **版权所有 © 2024 OpenDILab。保留所有权利。**
     """
 
 # 路径变量，方便之后的文件使用
-file_path = './documents/LightZero_README.zh.md'
+file_path = './documents/LightZero_README_zh.md'
 
 # 加载原始Markdown文档
 loader = TextLoader(file_path)
@@ -42,21 +49,19 @@ orig_documents = loader.load()
 conversation_history = []
 
 
-def rag_answer(question, model_name, temperature, embedding_model, k):
+def rag_answer(question, temperature, k):
     """
     处理用户问题并返回答案和高亮显示的上下文
 
     :param question: 用户输入的问题
-    :param model_name: 使用的语言模型名称
     :param temperature: 生成答案时使用的温度参数
-    :param embedding_model: 使用的嵌入模型
     :param k: 检索到的文档块数量
     :return: 模型生成的答案和高亮显示上下文的Markdown文本
     """
     try:
         chunks = load_and_split_document(file_path, chunk_size=5000, chunk_overlap=500)
-        retriever = create_vector_store(chunks, model=embedding_model, k=k)
-        rag_chain = setup_rag_chain(model_name=model_name, temperature=temperature)
+        retriever = create_vector_store(chunks, model='OpenAI', k=k)
+        rag_chain = setup_rag_chain(model_name='kimi', temperature=temperature)
 
         # 将问题添加到对话历史中
         conversation_history.append(("User", question))
@@ -64,8 +69,9 @@ def rag_answer(question, model_name, temperature, embedding_model, k):
         # 将对话历史转换为字符串
         history_str = "\n".join([f"{role}: {text}" for role, text in conversation_history])
 
-        retrieved_documents, answer = execute_query(retriever, rag_chain, history_str, model_name=model_name,
+        retrieved_documents, answer = execute_query(retriever, rag_chain, history_str, model_name='kimi',
                                                     temperature=temperature)
+
         # 在文档中高亮显示上下文
         context = [retrieved_documents[i].page_content for i in range(len(retrieved_documents))]
         highlighted_document = orig_documents[0].page_content
@@ -74,10 +80,17 @@ def rag_answer(question, model_name, temperature, embedding_model, k):
 
         # 将回答添加到对话历史中
         conversation_history.append(("Assistant", answer))
+
+        # 将对话历史存储到数据库中（此处省略数据库操作代码）
+
+        # 返回完整的对话历史
+        full_history = "\n".join([f"{role}: {text}" for role, text in conversation_history])
+
     except Exception as e:
         print(f"An error occurred: {e}")
-        return "处理您的问题时出现错误，请稍后再试。", ""
-    return answer, highlighted_document
+        return "处理您的问题时出现错误，请稍后再试。", "", ""
+
+    return answer, highlighted_document, full_history
 
 
 def clear_context():
@@ -86,28 +99,28 @@ def clear_context():
     """
     global conversation_history
     conversation_history = []
-    return "", ""
+    return "", "", ""
+
+
+def export_history():
+    """
+    导出对话历史记录
+    """
+    # 从数据库中获取完整的对话历史记录（此处省略数据库操作代码）
+    exported_history = "对话历史记录：\n" + "\n".join([f"{role}: {text}" for role, text in conversation_history])
+    return exported_history
 
 
 if __name__ == "__main__":
-    with gr.Blocks(title=title, theme='ParityError/Interstellar') as rag_demo:
+    with gr.Blocks(title=title, theme='ParityError/Interstellar') as zero_pal:
         gr.Markdown(title_markdown)
 
         with gr.Row():
             with gr.Column():
                 inputs = gr.Textbox(
-                    placeholder="请您输入任何关于 LightZero 的问题。",
+                    placeholder="请您在这里输入任何关于 LightZero 的问题。",
                     label="问题 (Q)")
-                model_name = gr.Dropdown(
-                    choices=['kimi', 'abab6-chat', 'glm-4', 'gpt-3.5-turbo', 'gpt-4', 'gpt-4-turbo', 'azure_gpt-4', 'azure_gpt-35-turbo-16k', 'azure_gpt-35-turbo'],
-                    # value='azure_gpt-4',
-                    value='kimi',
-                    label="选择语言模型")
                 temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.01, step=0.01, label="温度参数")
-                embedding_model = gr.Dropdown(
-                    choices=['HuggingFace', 'TensorflowHub', 'OpenAI'],
-                    value='OpenAI',
-                    label="选择嵌入模型")
                 k = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="检索到的文档块数量")
                 with gr.Row():
                     gr_submit = gr.Button('提交')
@@ -115,18 +128,17 @@ if __name__ == "__main__":
 
             outputs_answer = gr.Textbox(placeholder="当你点击提交按钮后，这里会显示 RAG 模型给出的回答。",
                                         label="回答 (A)")
+        outputs_history = gr.Textbox(label="对话历史")
         with gr.Row():
             outputs_context = gr.Markdown(label="参考的文档，检索得到的 context 用高亮显示 (C)")
-
-        gr.Markdown(tos_markdown)
-
+        gr_clear.click(clear_context, outputs=[outputs_context, outputs_history])
         gr_submit.click(
             rag_answer,
-            inputs=[inputs, model_name, temperature, embedding_model, k],
-            outputs=[outputs_answer, outputs_context],
+            inputs=[inputs, temperature, k],
+            outputs=[outputs_answer, outputs_context, outputs_history],
         )
-        gr_clear.click(clear_context, outputs=[outputs_answer, outputs_context])
+        gr.Markdown(tos_markdown)
 
     concurrency = int(os.environ.get('CONCURRENCY', os.cpu_count()))
     favicon_path = os.path.join(os.path.dirname(__file__), 'assets', 'avatar.png')
-    rag_demo.queue().launch(max_threads=concurrency, favicon_path=favicon_path, share=True)
+    zero_pal.queue().launch(max_threads=concurrency, favicon_path=favicon_path, share=True)

app_mqa_database.py

@@ -0,0 +1,214 @@
+import os
+import sqlite3
+import threading
+
+import gradio as gr
+from dotenv import load_dotenv
+from langchain.document_loaders import TextLoader
+
+from RAG.analyze_conversation_history import analyze_conversation_history
+from rag_demo import load_and_split_document, create_vector_store, setup_rag_chain, execute_query
+
+# 环境设置
+load_dotenv()  # 加载环境变量
+QUESTION_LANG = os.getenv("QUESTION_LANG")  # 从环境变量获取 QUESTION_LANG
+assert QUESTION_LANG in ['cn', 'en'], QUESTION_LANG
+
+if QUESTION_LANG == "cn":
+    title = "ZeroPal"
+    title_markdown = """
+    <div align="center">
+        <img src="https://raw.githubusercontent.com/puyuan1996/RAG/main/assets/banner.svg" width="80%" height="20%" alt="Banner Image">
+    </div>
+    
+    📢 **操作说明**：请在下方的“问题”框中输入关于 LightZero 的问题，并点击“提交”按钮。右侧的“回答”框将展示 RAG 模型提供的答案。
+    您可以在问答框下方查看当前“对话历史”，点击“清除上下文”按钮可清空历史记录。在“对话历史”框下方，您将找到相关参考文档，其中相关文段将以黄色高亮显示。
+    如果您喜欢这个项目，请在 GitHub [LightZero RAG Demo](https://github.com/puyuan1996/RAG) 上给我们点赞！✨ 您的支持是我们持续更新的动力。
+    
+    <div align="center">
+        <strong>注意：算法模型输出可能包含一定的随机性。结果不代表开发者和相关 AI 服务的态度和意见。本项目开发者不对结果作出任何保证，仅供参考之用。使用该服务即代表同意后文所述的使用条款。</strong>
+    </div>
+    """
+    tos_markdown = """
+    ### 使用条款
+    
+    使用本服务的玩家需同意以下条款：
+    
+    - 本服务为探索性研究的预览版，仅供非商业用途。
+    - 服务不得用于任何非法、有害、暴力、种族主义或其他令人反感的目的。
+    - 服务提供有限的安全措施，并可能生成令人反感的内容。
+    - 如果您对服务体验不满，请通过 [email protected] 与我们联系！我们承诺修复问题并不断改进项目。
+    - 为了获得最佳体验，请使用台式电脑，因为移动设备可能会影响视觉效果。
+    
+    **版权所有 © 2024 OpenDILab。保留所有权利。**
+    """
+
+# 路径变量,方便之后的文件使用
+file_path = './documents/LightZero_README_zh.md'
+
+# 加载原始Markdown文档
+loader = TextLoader(file_path)
+orig_documents = loader.load()
+
+# 存储对话历史
+conversation_history = {}
+
+# 创建线程局部数据对象
+threadLocal = threading.local()
+
+
+def get_db_connection():
+    """
+    返回当前线程的数据库连接
+    """
+    conn = getattr(threadLocal, 'conn', None)
+    if conn is None:
+        # 连接到SQLite数据库
+        conn = sqlite3.connect('database/conversation_history.db')
+        c = conn.cursor()
+        # Drop the existing 'history' table if it exists
+        # c.execute('DROP TABLE IF EXISTS history')
+        # 创建存储对话历史的表
+        c.execute('''CREATE TABLE IF NOT EXISTS history
+                     (id INTEGER PRIMARY KEY AUTOINCREMENT,
+                     user_id TEXT NOT NULL,
+                     user_input TEXT NOT NULL,
+                     assistant_output TEXT NOT NULL,
+                     timestamp DATETIME DEFAULT CURRENT_TIMESTAMP)''')
+        threadLocal.conn = conn
+    return conn
+
+
+def get_db_cursor():
+    """
+    返回当前线程的数据库游标
+    """
+    conn = get_db_connection()
+    c = getattr(threadLocal, 'cursor', None)
+    if c is None:
+        c = conn.cursor()
+        threadLocal.cursor = c
+    return c
+
+
+# 程序结束时清理数据库连接
+def close_db_connection():
+    conn = getattr(threadLocal, 'conn', None)
+    if conn is not None:
+        conn.close()
+        setattr(threadLocal, 'conn', None)
+
+    c = getattr(threadLocal, 'cursor', None)
+    if c is not None:
+        c.close()
+        setattr(threadLocal, 'cursor', None)
+
+
+def rag_answer(question, temperature, k, user_id):
+    """
+    处理用户问题并返回答案和高亮显示的上下文
+
+    :param question: 用户输入的问题
+    :param temperature: 生成答案时使用的温度参数
+    :param k: 检索到的文档块数量
+    :param user_id: 用户ID
+    :return: 模型生成的答案和高亮显示上下文的Markdown文本
+    """
+    try:
+        chunks = load_and_split_document(file_path, chunk_size=5000, chunk_overlap=500)
+        retriever = create_vector_store(chunks, model='OpenAI', k=k)
+        rag_chain = setup_rag_chain(model_name='kimi', temperature=temperature)
+
+        if user_id not in conversation_history:
+            conversation_history[user_id] = []
+
+        conversation_history[user_id].append((f"User[{user_id}]", question))
+
+        history_str = "\n".join([f"{role}: {text}" for role, text in conversation_history[user_id]])
+
+        retrieved_documents, answer = execute_query(retriever, rag_chain, history_str, model_name='kimi',
+                                                    temperature=temperature)
+
+        ############################
+        # 获取当前线程的数据库连接和游标
+        ############################
+        conn = get_db_connection()
+        c = get_db_cursor()
+
+        # 分析对话历史
+        # analyze_conversation_history()
+        # 获取总的对话记录数
+        c.execute("SELECT COUNT(*) FROM history")
+        total_records = c.fetchone()[0]
+        print(f"总对话记录数: {total_records}")
+
+        # 将问题和回答存储到数据库
+        c.execute("INSERT INTO history (user_id, user_input, assistant_output) VALUES (?, ?, ?)",
+                  (user_id, question, answer))
+        conn.commit()
+
+        # 在文档中高亮显示上下文
+        context = [retrieved_documents[i].page_content for i in range(len(retrieved_documents))]
+        highlighted_document = orig_documents[0].page_content
+        for i in range(len(context)):
+            highlighted_document = highlighted_document.replace(context[i], f"<mark>{context[i]}</mark>")
+
+        conversation_history[user_id].append(("Assistant", answer))
+
+        full_history = "\n".join([f"{role}: {text}" for role, text in conversation_history[user_id]])
+    except Exception as e:
+        print(f"An error occurred: {e}")
+        return "处理您的问题时出现错误,请稍后再试。", "", ""
+    finally:
+        # 不再在这里关闭游标和连接
+        pass
+
+    return answer, highlighted_document, full_history
+
+
+def clear_context(user_id):
+    """
+    清除对话历史
+    """
+    if user_id in conversation_history:
+        conversation_history[user_id] = []
+    return "", "", ""
+
+
+if __name__ == "__main__":
+    with gr.Blocks(title=title, theme='ParityError/Interstellar') as zero_pal:
+        gr.Markdown(title_markdown)
+
+        with gr.Row():
+            with gr.Column():
+                user_id = gr.Textbox(
+                    placeholder="请输入您的真实姓名或昵称作为用户ID",
+                    label="用户ID")
+                inputs = gr.Textbox(
+                    placeholder="请您在这里输入任何关于 LightZero 的问题。",
+                    label="问题")
+                temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.01, step=0.01, label="温度参数")
+                k = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="检索到的文档块数量")
+                with gr.Row():
+                    gr_submit = gr.Button('提交')
+                    gr_clear = gr.Button('清除上下文')
+
+            outputs_answer = gr.Textbox(placeholder="当你点击提交按钮后,这里会显示 RAG 模型给出的回答。",
+                                        label="回答")
+        outputs_history = gr.Textbox(label="对话历史")
+        with gr.Row():
+            outputs_context = gr.Markdown(label="参考的文档(检索得到的相关文段用高亮显示)")
+        gr_clear.click(clear_context, inputs=user_id, outputs=[outputs_context, outputs_history])
+        gr_submit.click(
+            rag_answer,
+            inputs=[inputs, temperature, k, user_id],
+            outputs=[outputs_answer, outputs_context, outputs_history],
+        )
+        gr.Markdown(tos_markdown)
+
+    concurrency = int(os.environ.get('CONCURRENCY', os.cpu_count()))
+    favicon_path = os.path.join(os.path.dirname(__file__), 'assets', 'avatar.png')
+    zero_pal.queue().launch(max_threads=concurrency, favicon_path=favicon_path, share=True)
+
+    # 在合适的地方，例如程序退出时，调用close_db_connection函数
+    close_db_connection()

app_qa.py

@@ -1,106 +0,0 @@
-import os
-
-import gradio as gr
-from dotenv import load_dotenv
-from langchain.document_loaders import TextLoader
-
-from rag_demo import load_and_split_document, create_vector_store, setup_rag_chain, execute_query
-
-# 环境设置
-load_dotenv()  # 加载环境变量
-QUESTION_LANG = os.getenv("QUESTION_LANG")  # 从环境变量获取 QUESTION_LANG
-assert QUESTION_LANG in ['cn', 'en'], QUESTION_LANG
-
-if QUESTION_LANG == "cn":
-    title = "LightZero RAG Demo"
-    title_markdown = """
-    <div align="center">
-        <img src="https://raw.githubusercontent.com/puyuan1996/RAG/main/assets/banner.svg" width="80%" height="20%" alt="Banner Image">
-    </div>
-    <h2 style="text-align: center; color: black;"><a href="https://github.com/puyuan1996/RAG"> LightZero RAG Demo</a></h2>
-    <h4 align="center"> 📢说明：请您在下面的"问题（Q）"框中输入任何关于 LightZero 的问题，然后点击"提交"按钮。右侧"回答（A）"框中会显示 RAG 模型给出的回答。在 QA 栏的下方会给出参考文档（其中检索得到的相关文段会用黄色高亮显示）。</h4>
-    <h4 align="center"> 如果你喜欢这个项目，请给我们在 GitHub 点个 star ✨ 。我们将会持续保持更新。  </h4>
-    <strong><h5 align="center">注意：算法模型的输出可能包含一定的随机性。相关结果不代表任何开发者和相关 AI 服务的态度和意见。本项目开发者不对生成结果作任何保证，仅供参考。<h5></strong>
-    """
-    tos_markdown = """
-    ### 使用条款
-    玩家使用本服务须同意以下条款：
-    该服务是一项探索性研究预览版，仅供非商业用途。它仅提供有限的安全措施，并可能生成令人反感的内容。不得将其用于任何非法、有害、暴力、种族主义等目的。
-    如果您的游玩体验有不佳之处，请发送邮件至 [email protected] ！ 我们将删除相关信息，并不断改进这个项目。
-    为了获得最佳体验，请使用台式电脑，因为移动设备可能会影响可视化效果。
-    **版权所有 2024 OpenDILab。**
-    """
-
-# 路径变量，方便之后的文件使用
-file_path = './documents/LightZero_README.zh.md'
-
-# 加载原始Markdown文档
-loader = TextLoader(file_path)
-orig_documents = loader.load()
-
-def rag_answer(question, model_name, temperature, embedding_model, k):
-    """
-    处理用户问题并返回答案和高亮显示的上下文
-
-    :param question: 用户输入的问题
-    :param model_name: 使用的语言模型名称
-    :param temperature: 生成答案时使用的温度参数
-    :param embedding_model: 使用的嵌入模型
-    :param k: 检索到的文档块数量
-    :return: 模型生成的答案和高亮显示上下文的Markdown文本
-    """
-    try:
-        chunks = load_and_split_document(file_path, chunk_size=5000, chunk_overlap=500)
-        retriever = create_vector_store(chunks, model=embedding_model, k=k)
-        rag_chain = setup_rag_chain(model_name=model_name, temperature=temperature)
-
-        retrieved_documents, answer = execute_query(retriever, rag_chain, question, model_name=model_name, temperature=temperature)
-        # 在文档中高亮显示上下文
-        context = [retrieved_documents[i].page_content for i in range(len(retrieved_documents))]
-        highlighted_document = orig_documents[0].page_content
-        for i in range(len(context)):
-            highlighted_document = highlighted_document.replace(context[i], f"<mark>{context[i]}</mark>")
-    except Exception as e:
-        print(f"An error occurred: {e}")
-        return "处理您的问题时出现错误，请稍后再试。", ""
-    return answer, highlighted_document
-
-
-if __name__ == "__main__":
-    with gr.Blocks(title=title, theme='ParityError/Interstellar') as rag_demo:
-        gr.Markdown(title_markdown)
-
-        with gr.Row():
-            with gr.Column():
-                inputs = gr.Textbox(
-                    placeholder="请您输入任何关于 LightZero 的问题。",
-                    label="问题 (Q)")
-                model_name = gr.Dropdown(
-                    choices=['kimi', 'abab6-chat', 'glm-4', 'gpt-3.5-turbo', 'gpt-4', 'gpt-4-turbo', 'azure_gpt-4', 'azure_gpt-35-turbo-16k', 'azure_gpt-35-turbo'],
-                    # value='azure_gpt-4',
-                    value='kimi',
-                    label="选择语言模型")
-                temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.01, step=0.01, label="温度参数")
-                embedding_model = gr.Dropdown(
-                    choices=['HuggingFace', 'TensorflowHub', 'OpenAI'],
-                    value='OpenAI',
-                    label="选择嵌入模型")
-                k = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="检索到的文档块数量")
-                gr_submit = gr.Button('提交')
-
-            outputs_answer = gr.Textbox(placeholder="当你点击提交按钮后，这里会显示 RAG 模型给出的回答。",
-                                        label="回答 (A)")
-        with gr.Row():
-            outputs_context = gr.Markdown(label="参考的文档，检索得到的 context 用高亮显示 (C)")
-
-        gr.Markdown(tos_markdown)
-
-        gr_submit.click(
-            rag_answer,
-            inputs=[inputs, model_name, temperature, embedding_model, k],
-            outputs=[outputs_answer, outputs_context],
-        )
-
-    concurrency = int(os.environ.get('CONCURRENCY', os.cpu_count()))
-    favicon_path = os.path.join(os.path.dirname(__file__), 'assets', 'avatar.png')
-    rag_demo.queue().launch(max_threads=concurrency, favicon_path=favicon_path, share=True)

documents/LightZero_README.md

@@ -27,7 +27,7 @@
 [![Contributors](https://img.shields.io/github/contributors/opendilab/LightZero)](https://github.com/opendilab/LightZero/graphs/contributors)
 [![GitHub license](https://img.shields.io/github/license/opendilab/LightZero)](https://github.com/opendilab/LightZero/blob/master/LICENSE)
 
-Updated on 2023.12.07 LightZero-v0.0.3
+Updated on 2024.03.15 LightZero-v0.0.4
 
 > LightZero is a lightweight, efficient, and easy-to-understand open-source algorithm toolkit that combines Monte Carlo Tree Search (MCTS) and Deep Reinforcement Learning (RL). 
 
@@ -207,6 +207,15 @@ cd LightZero
 python3 -u zoo/board_games/tictactoe/config/tictactoe_muzero_bot_mode_config.py
 ```
 
+## Customization Documentation
+
+For those looking to tailor environments and algorithms, we offer comprehensive guides:
+
+- **Environments:** [Customize Environments](https://github.com/opendilab/LightZero/blob/main/docs/source/tutorials/envs/customize_envs.md)
+- **Algorithms:** [Customize Algorithms](https://github.com/opendilab/LightZero/blob/main/docs/source/tutorials/algos/customize_algos.md)
+
+Should you have any questions, feel free to contact us for support.
+
 ## Benchmark
 
 <details open><summary>Click to collapse</summary>
@@ -374,6 +383,14 @@ Here is a collection of research papers about **Monte Carlo Tree Search**.
   - ExpEnv: USPTO datasets
   - [Code](https://github.com/binghong-ml/retro_star)
 #### ICLR
+- [The Update Equivalence Framework for Decision-Time Planning](https://openreview.net/forum?id=JXGph215fL) 2024
+  - Samuel Sokota, Gabriele Farina, David J Wu, Hengyuan Hu, Kevin A. Wang, J Zico Kolter, Noam Brown
+  - Key: imperfect-information games, search, decision-time planning, update equivalence
+  - ExpEnv: Hanabi, 3x3 Abrupt Dark Hex and Phantom Tic-Tac-Toe
+- [Efficient Multi-agent Reinforcement Learning by Planning](https://openreview.net/forum?id=CpnKq3UJwp) 2024
+  - Qihan Liu, Jianing Ye, Xiaoteng Ma, Jun Yang, Bin Liang, Chongjie Zhang
+  - Key: multi-agent reinforcement learning, planning, multi-agent MCTS
+  - ExpEnv: SMAC, LunarLander, MuJoCo, and Google Research Football
 - [Become a Proficient Player with Limited Data through Watching Pure Videos](https://openreview.net/pdf?id=Sy-o2N0hF4f) 2023
   - Weirui Ye, Yunsheng Zhang, Pieter Abbeel, Yang Gao
   - Key: pre-training from action-free videos, forward-inverse cycle consistency (FICC) objective based on vector quantization, pre-training phase, fine-tuning phase.
@@ -442,6 +459,10 @@ Here is a collection of research papers about **Monte Carlo Tree Search**.
   - Yangqing Fu, Ming Sun, Buqing Nie, Yue Gao
   - Key: probability tree state abstraction, transitivity and aggregation error bound
   - ExpEnv: Atari, CartPole, LunarLander, Gomoku
+- [Spending Thinking Time Wisely: Accelerating MCTS with Virtual Expansions](https://openreview.net/pdf?id=B_LdLljS842) 2022
+  - Weirui Ye, Pieter Abbeel, Yang Gao
+  - Key: trade off computation versus performancem, virtual expansions, spend thinking time adaptively.
+  - ExpEnv: Atari, 9x9 Go
 - [Planning for Sample Efficient Imitation Learning](https://openreview.net/forum?id=BkN5UoAqF7) 2022
   - Zhao-Heng Yin, Weirui Ye, Qifeng Chen, Yang Gao
   - Key: Behavioral Cloning，Adversarial Imitation Learning (AIL)，MCTS-based RL.
@@ -485,6 +506,7 @@ Here is a collection of research papers about **Monte Carlo Tree Search**.
   - [Code](https://github.com/matthewfaw/mixnmatch)
 
 #### Other Conference or Journal
+- [Learning to Stop: Dynamic Simulation Monte-Carlo Tree Search](https://arxiv.org/pdf/2012.07910.pdf) AAAI 2021.
 - [On Monte Carlo Tree Search and Reinforcement Learning](https://www.jair.org/index.php/jair/article/download/11099/26289/20632) Journal of Artificial Intelligence Research 2017.
 - [Sample-Efficient Neural Architecture Search by Learning Actions for Monte Carlo Tree Search](https://arxiv.org/pdf/1906.06832) IEEE Transactions on Pattern Analysis and Machine Intelligence 2022.
 </details>

documents/{LightZero_README.zh.md → LightZero_README_zh.md}

@@ -27,7 +27,7 @@
 [![Contributors](https://img.shields.io/github/contributors/opendilab/LightZero)](https://github.com/opendilab/LightZero/graphs/contributors)
 [![GitHub license](https://img.shields.io/github/license/opendilab/LightZero)](https://github.com/opendilab/LightZero/blob/master/LICENSE)
 
-最近更新于 2023.12.07 LightZero-v0.0.3
+最近更新于 2024.03.15 LightZero-v0.0.4
 
 > LightZero 是一个轻量、高效、易懂的 MCTS+RL 开源算法库。
 
@@ -191,6 +191,14 @@ python3 -u zoo/atari/config/atari_muzero_config.py
 cd LightZero
 python3 -u zoo/board_games/tictactoe/config/tictactoe_muzero_bot_mode_config.py
 ```
+## 定制化文档
+
+为希望定制环境和算法的用户，我们提供了全面的指南：
+
+- **环境定制：** [定制环境](https://github.com/opendilab/LightZero/blob/main/docs/source/tutorials/envs/customize_envs_zh.md)
+- **算法定制：** [定制算法](https://github.com/opendilab/LightZero/blob/main/docs/source/tutorials/algos/customize_algos_zh.md)
+
+如有任何疑问，欢迎随时联系我们寻求帮助。
 
 ## 基线算法比较
 
@@ -352,7 +360,7 @@ python3 -u zoo/board_games/tictactoe/config/tictactoe_muzero_bot_mode_config.py
   - ExpEnv: Gridworld and SysAdmin
 - [Efficient Learning for AlphaZero via Path Consistency](https://proceedings.mlr.press/v162/zhao22h/zhao22h.pdf) 2022
   - Dengwei Zhao, Shikui Tu, Lei Xu
-  - Key: limited amount of self-plays,  path consistency (PC) optimality
+  - Key: limited amount of self-plays, path consistency (PC) optimality
   - ExpEnv: Go, Othello, Gomoku
 - [Visualizing MuZero Models](https://arxiv.org/abs/2102.12924) 2021
   - Joery A. de Vries, Ken S. Voskuil, Thomas M. Moerland, Aske Plaat
@@ -361,7 +369,7 @@ python3 -u zoo/board_games/tictactoe/config/tictactoe_muzero_bot_mode_config.py
 and internal state transition dynamics,
 - [Convex Regularization in Monte-Carlo Tree Search](https://arxiv.org/pdf/2007.00391.pdf) 2021
   - Tuan Dam, Carlo D'Eramo, Jan Peters, Joni Pajarinen
-  - Key: entropy-regularization backup operators, regret analysis, Tsallis etropy, 
+  - Key: entropy-regularization backup operators, regret analysis, Tsallis etropy
   - ExpEnv: synthetic tree, Atari
 - [Information Particle Filter Tree: An Online Algorithm for POMDPs with Belief-Based Rewards on Continuous Domains](http://proceedings.mlr.press/v119/fischer20a/fischer20a.pdf) 2020
   - Johannes Fischer, Ömer Sahin Tas
@@ -374,6 +382,14 @@ and internal state transition dynamics,
   - ExpEnv: USPTO datasets
   - [Code](https://github.com/binghong-ml/retro_star)
 #### ICLR
+- [The Update Equivalence Framework for Decision-Time Planning](https://openreview.net/forum?id=JXGph215fL) 2024
+  - Samuel Sokota, Gabriele Farina, David J Wu, Hengyuan Hu, Kevin A. Wang, J Zico Kolter, Noam Brown
+  - Key: imperfect-information games, search, decision-time planning, update equivalence
+  - ExpEnv: Hanabi, 3x3 Abrupt Dark Hex and Phantom Tic-Tac-Toe
+- [Efficient Multi-agent Reinforcement Learning by Planning](https://openreview.net/forum?id=CpnKq3UJwp) 2024
+  - Qihan Liu, Jianing Ye, Xiaoteng Ma, Jun Yang, Bin Liang, Chongjie Zhang
+  - Key: multi-agent reinforcement learning, planning, multi-agent MCTS
+  - ExpEnv: SMAC, LunarLander, MuJoCo, and Google Research Football
 - [Become a Proficient Player with Limited Data through Watching Pure Videos](https://openreview.net/pdf?id=Sy-o2N0hF4f) 2023
   - Weirui Ye, Yunsheng Zhang, Pieter Abbeel, Yang Gao
   - Key: pre-training from action-free videos, forward-inverse cycle consistency (FICC) objective based on vector quantization, pre-training phase, fine-tuning phase.
@@ -421,8 +437,8 @@ and internal state transition dynamics,
   - Binghong Chen, Bo Dai, Qinjie Lin, Guo Ye, Han Liu, Le Song
   - Key: meta path planning algorithm, exploits a novel neural architecture which can learn promising search directions from problem structures.
   - ExpEnv: a 2d workspace with a 2 DoF (degrees of freedom) point robot, a 3 DoF stick robot and a 5 DoF snake robot
-#### NeurIPS
 
+#### NeurIPS
 - [LightZero: A Unified Benchmark for Monte Carlo Tree Search in General Sequential Decision Scenarios](https://openreview.net/pdf?id=oIUXpBnyjv) 2023
   - Yazhe Niu, Yuan Pu, Zhenjie Yang, Xueyan Li, Tong Zhou, Jiyuan Ren, Shuai Hu, Hongsheng Li, Yu Liu
   - Key: the first unified benchmark for deploying MCTS/MuZero in general sequential decision scenarios.
@@ -443,6 +459,10 @@ and internal state transition dynamics,
   - Yangqing Fu, Ming Sun, Buqing Nie, Yue Gao
   - Key: probability tree state abstraction, transitivity and aggregation error bound
   - ExpEnv: Atari, CartPole, LunarLander, Gomoku
+- [Spending Thinking Time Wisely: Accelerating MCTS with Virtual Expansions](https://openreview.net/pdf?id=B_LdLljS842) 2022
+  - Weirui Ye, Pieter Abbeel, Yang Gao
+  - Key: trade off computation versus performancem, virtual expansions, spend thinking time adaptively.
+  - ExpEnv: Atari, 9x9 Go
 - [Planning for Sample Efficient Imitation Learning](https://openreview.net/forum?id=BkN5UoAqF7) 2022
   - Zhao-Heng Yin, Weirui Ye, Qifeng Chen, Yang Gao
   - Key: Behavioral Cloning，Adversarial Imitation Learning (AIL)，MCTS-based RL，
@@ -486,6 +506,7 @@ and internal state transition dynamics,
   - [Code](https://github.com/matthewfaw/mixnmatch)
 
 #### Other Conference or Journal
+- [Learning to Stop: Dynamic Simulation Monte-Carlo Tree Search](https://arxiv.org/pdf/2012.07910.pdf) AAAI 2021.
 - [On Monte Carlo Tree Search and Reinforcement Learning](https://www.jair.org/index.php/jair/article/download/11099/26289/20632) Journal of Artificial Intelligence Research 2017.
 - [Sample-Efficient Neural Architecture Search by Learning Actions for Monte Carlo Tree Search](https://arxiv.org/pdf/1906.06832) IEEE Transactions on Pattern Analysis and Machine Intelligence 2022.
 </details>

rag_demo.py

@@ -234,11 +234,11 @@ def execute_query_no_rag(model_name="gpt-4", temperature=0, query=""):
 
 if __name__ == "__main__":
     # 假设文档已存在于本地
-    file_path = './documents/LightZero_README.zh.md'
+    file_path = './documents/LightZero_README_zh.md'
     # model_name = "glm-4"  # model_name=['abab6-chat', 'glm-4', 'gpt-3.5-turbo', 'gpt-4', 'gpt-4-turbo', 'azure_gpt-4', 'azure_gpt-35-turbo-16k', 'azure_gpt-35-turbo']
-    model_name = 'azure_gpt-4'
+    # model_name = 'azure_gpt-4'
+    model_name = 'kimi'
     temperature = 0.01
-    # embedding_model = 'HuggingFace'  # embedding_model=['HuggingFace', 'TensorflowHub', 'OpenAI']
     embedding_model = 'OpenAI'  # embedding_model=['HuggingFace', 'TensorflowHub', 'OpenAI']
 
     # 加载和分割文档
@@ -251,11 +251,11 @@ if __name__ == "__main__":
     rag_chain = setup_rag_chain(model_name=model_name, temperature=temperature)
 
     # 提出问题并获取答案
-    query = ("GitHub - opendilab/LightZero: [NeurIPS 2023 Spotlight] LightZero: A Unified Benchmark for Monte Carl  请根据这个仓库回答下面的问题：（1）请简要介绍一下 LightZero （2）请详细介绍 LightZero 的框架结构。 （3）请给出安装 LightZero，运行他们的示例代码的详细步骤 （4）- 请问 LightZero 具体支持什么任务（tasks/environments）? （5）请问 LightZero 具体支持什么算法?（6）请问 LightZero 具体支持什么算法，各自支持在哪些任务上运行? （7）请问 LightZero 里面实现的 MuZero 算法支持在 Atari 任务上运行吗？（8）请问 LightZero 里面实现的 AlphaZero 算法支持在 Atari 任务上运行吗？（9）LightZero 支持哪些算法? 各自的优缺点是什么? 我应该如何根据任务特点进行选择呢？（10）请结合 LightZero 中的代码介绍他们是如何实现 MCTS 的。（11）请问对这个仓库提出详细的改进建议")
+    query = ("请回答下面的问题：（1）请简要介绍一下 LightZero。（2）请详细介绍 LightZero 的框架结构。 （3）请给出安装 LightZero，运行他们的示例代码的详细步骤。（4）- 请问 LightZero 具体支持什么任务（tasks/environments）? （5）请问 LightZero 具体支持什么算法?（6）请问 LightZero 具体支持什么算法，各自支持在哪些任务上运行? （7）请问 LightZero 里面实现的 MuZero 算法支持在 Atari 任务上运行吗？（8）请问 LightZero 里面实现的 AlphaZero 算法支持在 Atari 任务上运行吗？（9）LightZero 支持哪些算法? 各自的优缺点是什么? 我应该如何根据任务特点进行选择呢？（10）请结合 LightZero 中的代码介绍他们是如何实现 MCTS 的。（11）请问对这个仓库提出详细的改进建议")
     """
-    （1）请简要介绍一下 LightZero 
+    （1）请简要介绍一下 LightZero。 
     （2）请详细介绍 LightZero 的框架结构。 
-    （3）请给出安装 LightZero，运行他们的示例代码的详细步骤 
+    （3）请给出安装 LightZero，运行他们的示例代码的详细步骤 。
     （4）请问 LightZero 具体支持什么任务（tasks/environments）? 
     （5）请问 LightZero 具体支持什么算法?
     （6）请问 LightZero 具体支持什么算法，各自支持在哪些任务上运行? 
@@ -266,6 +266,7 @@ if __name__ == "__main__":
     （11）请问对这个仓库提出详细的改进建议。
     """
 
+    # query = ("请检索最近关于Transformer+RL的最新论文，并给出详细介绍")
     # 使用 RAG 链获取参考的文档与答案
     retrieved_documents, result_with_rag = execute_query(retriever, rag_chain, query, model_name=model_name,
                                                          temperature=temperature)