webui.py 6.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181
  1. import gradio as gr
  2. import os
  3. import shutil
  4. from chains.local_doc_qa import LocalDocQA
  5. from configs.model_config import *
  6. # return top-k text chunk from vector store
  7. VECTOR_SEARCH_TOP_K = 6
  8. # LLM input history length
  9. LLM_HISTORY_LEN = 3
  10. def get_file_list():
  11. if not os.path.exists("content"):
  12. return []
  13. return [f for f in os.listdir("content")]
  14. file_list = get_file_list()
  15. embedding_model_dict_list = list(embedding_model_dict.keys())
  16. llm_model_dict_list = list(llm_model_dict.keys())
  17. local_doc_qa = LocalDocQA()
  18. def upload_file(file):
  19. if not os.path.exists("content"):
  20. os.mkdir("content")
  21. filename = os.path.basename(file.name)
  22. shutil.move(file.name, "content/" + filename)
  23. # file_list首位插入新上传的文件
  24. file_list.insert(0, filename)
  25. return gr.Dropdown.update(choices=file_list, value=filename)
  26. def get_answer(query, vs_path, history):
  27. if vs_path:
  28. resp, history = local_doc_qa.get_knowledge_based_answer(
  29. query=query, vs_path=vs_path, chat_history=history)
  30. else:
  31. history = history + [[None, "请先加载文件后,再进行提问。"]]
  32. return history, ""
  33. def update_status(history, status):
  34. history = history + [[None, status]]
  35. print(status)
  36. return history
  37. def init_model():
  38. try:
  39. local_doc_qa.init_cfg()
  40. return """模型已成功加载,请选择文件后点击"加载文件"按钮"""
  41. except Exception as e:
  42. print(e)
  43. return """模型未成功加载,请重新选择后点击"加载模型"按钮"""
  44. def reinit_model(llm_model, embedding_model, llm_history_len, top_k, history):
  45. try:
  46. local_doc_qa.init_cfg(llm_model=llm_model,
  47. embedding_model=embedding_model,
  48. llm_history_len=llm_history_len,
  49. top_k=top_k)
  50. model_status = """模型已成功重新加载,请选择文件后点击"加载文件"按钮"""
  51. except Exception as e:
  52. print(e)
  53. model_status = """模型未成功重新加载,请重新选择后点击"加载模型"按钮"""
  54. return history + [[None, model_status]]
  55. def get_vector_store(filepath, history):
  56. if local_doc_qa.llm and local_doc_qa.embeddings:
  57. vs_path = local_doc_qa.init_knowledge_vector_store(["content/" + filepath])
  58. if vs_path:
  59. file_status = "文件已成功加载,请开始提问"
  60. else:
  61. file_status = "文件未成功加载,请重新上传文件"
  62. else:
  63. file_status = "模型未完成加载,请先在加载模型后再导入文件"
  64. vs_path = None
  65. return vs_path, history + [[None, file_status]]
  66. block_css = """.importantButton {
  67. background: linear-gradient(45deg, #7e0570,#5d1c99, #6e00ff) !important;
  68. border: none !important;
  69. }
  70. .importantButton:hover {
  71. background: linear-gradient(45deg, #ff00e0,#8500ff, #6e00ff) !important;
  72. border: none !important;
  73. }"""
  74. webui_title = """
  75. # 🎉langchain-ChatGLM WebUI🎉
  76. 👍 [https://github.com/imClumsyPanda/langchain-ChatGLM](https://github.com/imClumsyPanda/langchain-ChatGLM)
  77. """
  78. init_message = """欢迎使用 langchain-ChatGLM Web UI,开始提问前,请依次如下 3 个步骤:
  79. 1. 选择语言模型、Embedding 模型及相关参数后点击"重新加载模型",并等待加载完成提示
  80. 2. 上传或选择已有文件作为本地知识文档输入后点击"重新加载文档",并等待加载完成提示
  81. 3. 输入要提交的问题后,点击回车提交 """
  82. model_status = init_model()
  83. with gr.Blocks(css=block_css) as demo:
  84. vs_path, file_status, model_status = gr.State(""), gr.State(""), gr.State(model_status)
  85. gr.Markdown(webui_title)
  86. with gr.Row():
  87. with gr.Column(scale=2):
  88. chatbot = gr.Chatbot([[None, init_message], [None, model_status.value]],
  89. elem_id="chat-box",
  90. show_label=False).style(height=750)
  91. query = gr.Textbox(show_label=False,
  92. placeholder="请输入提问内容,按回车进行提交",
  93. ).style(container=False)
  94. with gr.Column(scale=1):
  95. llm_model = gr.Radio(llm_model_dict_list,
  96. label="LLM 模型",
  97. value=LLM_MODEL,
  98. interactive=True)
  99. llm_history_len = gr.Slider(0,
  100. 10,
  101. value=LLM_HISTORY_LEN,
  102. step=1,
  103. label="LLM history len",
  104. interactive=True)
  105. embedding_model = gr.Radio(embedding_model_dict_list,
  106. label="Embedding 模型",
  107. value=EMBEDDING_MODEL,
  108. interactive=True)
  109. top_k = gr.Slider(1,
  110. 20,
  111. value=VECTOR_SEARCH_TOP_K,
  112. step=1,
  113. label="向量匹配 top k",
  114. interactive=True)
  115. load_model_button = gr.Button("重新加载模型")
  116. # with gr.Column():
  117. with gr.Tab("select"):
  118. selectFile = gr.Dropdown(file_list,
  119. label="content file",
  120. interactive=True,
  121. value=file_list[0] if len(file_list) > 0 else None)
  122. with gr.Tab("upload"):
  123. file = gr.File(label="content file",
  124. file_types=['.txt', '.md', '.docx', '.pdf']
  125. ) # .style(height=100)
  126. load_file_button = gr.Button("加载文件")
  127. load_model_button.click(reinit_model,
  128. show_progress=True,
  129. inputs=[llm_model, embedding_model, llm_history_len, top_k, chatbot],
  130. outputs=chatbot
  131. )
  132. # 将上传的文件保存到content文件夹下,并更新下拉框
  133. file.upload(upload_file,
  134. inputs=file,
  135. outputs=selectFile)
  136. load_file_button.click(get_vector_store,
  137. show_progress=True,
  138. inputs=[selectFile, chatbot],
  139. outputs=[vs_path, chatbot],
  140. )
  141. query.submit(get_answer,
  142. [query, vs_path, chatbot],
  143. [chatbot, query],
  144. )
  145. demo.queue(concurrency_count=3).launch(
  146. server_name='0.0.0.0', share=False, inbrowser=False)