加入川虎助理和川虎助理Pro模型

ChuanhuChatbot.py

@@ -286,7 +286,6 @@ with gr.Blocks(css=customCSS, theme=small_and_beautiful_theme) as demo:
             chatbot,
             use_streaming_checkbox,
             use_websearch_checkbox,
-            autogpt_mode,
             index_files,
             language_select_dropdown,
         ],
@@ -349,7 +348,6 @@ with gr.Blocks(css=customCSS, theme=small_and_beautiful_theme) as demo:
             chatbot,
             use_streaming_checkbox,
             use_websearch_checkbox,
-            autogpt_mode,
             index_files,
             language_select_dropdown,
         ],

config_example.json

@@ -15,6 +15,14 @@
     "local_embedding": false, //是否在本地编制索引
     "hide_history_when_not_logged_in": false, //未登录情况下是否不展示对话历史
     "default_model": "gpt-3.5-turbo", // 默认模型
+
+    //川虎助理设置
+    "default_chuanhu_assistant_model": "gpt-4", //川虎助理使用的模型，可选gpt-3.5或者gpt-4
+    "GOOGLE_CSE_ID": "", //谷歌搜索引擎ID，用于川虎助理Pro模式，获取方式请看 https://stackoverflow.com/questions/37083058/programmatically-searching-google-in-python-using-custom-search
+    "GOOGLE_API_KEY": "", //谷歌API Key，用于川虎助理Pro模式
+    "WOLFRAM_ALPHA_APPID": "", //Wolfram Alpha API Key，用于川虎助理Pro模式，获取方式请看 https://products.wolframalpha.com/api/
+    "SERPAPI_API_KEY": "", //SerpAPI API Key，用于川虎助理Pro模式，获取方式请看 https://serpapi.com/
+
     "advance_docs": {
         "pdf": {
             // 是否认为PDF是双栏的

modules/config.py

@@ -24,7 +24,8 @@ __all__ = [
     "server_name",
     "server_port",
     "share",
-    "hide_history_when_not_logged_in"
+    "hide_history_when_not_logged_in",
+    "default_chuanhu_assistant_model"
 ]
 
 # 添加一个统一的config文件，避免文件过多造成的疑惑（优先级最低）
@@ -102,6 +103,12 @@ api_host = os.environ.get("api_host", config.get("api_host", ""))
 if api_host:
     shared.state.set_api_host(api_host)
 
+default_chuanhu_assistant_model = config.get("default_chuanhu_assistant_model", "gpt-4")
+os.environ["GOOGLE_CSE_ID"] = config.get("GOOGLE_CSE_ID", "")
+os.environ["GOOGLE_API_KEY"] = config.get("GOOGLE_API_KEY", "")
+os.environ["WOLFRAM_ALPHA_APPID"] = config.get("WOLFRAM_ALPHA_APPID", "")
+os.environ["SERPAPI_API_KEY"] = config.get("SERPAPI_API_KEY", "")
+
 @contextmanager
 def retrieve_openai_api(api_key = None):
     old_api_key = os.environ.get("OPENAI_API_KEY", "")

modules/models/ChuanhuAgent.py

@@ -0,0 +1,108 @@
+from langchain.chains.summarize import load_summarize_chain
+from langchain import OpenAI, PromptTemplate, LLMChain
+from langchain.chat_models import ChatOpenAI
+from langchain.text_splitter import CharacterTextSplitter
+from langchain.chains.mapreduce import MapReduceChain
+from langchain.prompts import PromptTemplate
+from langchain.text_splitter import TokenTextSplitter
+from langchain.embeddings import OpenAIEmbeddings
+from langchain.vectorstores import FAISS
+from langchain.chains import RetrievalQA
+from langchain.agents import load_tools
+from langchain.agents import initialize_agent
+from langchain.agents import AgentType
+from langchain.docstore.document import Document
+from langchain.tools import BaseTool, StructuredTool, Tool, tool
+from langchain.callbacks.stdout import StdOutCallbackHandler
+from langchain.callbacks.manager import BaseCallbackManager
+
+from pydantic import BaseModel, Field
+
+import requests
+from bs4 import BeautifulSoup
+
+from .base_model import BaseLLMModel
+from ..config import default_chuanhu_assistant_model
+from ..presets import SUMMARIZE_PROMPT
+import logging
+
+class WebBrowsingInput(BaseModel):
+    url: str = Field(description="URL of a webpage")
+
+class WebAskingInput(BaseModel):
+    url: str = Field(description="URL of a webpage")
+    question: str = Field(description="Question that you want to know the answer to, based on the webpage's content.")
+
+
+class ChuanhuAgent_Client(BaseLLMModel):
+    def __init__(self, model_name, openai_api_key, user_name="") -> None:
+        super().__init__(model_name=model_name, user=user_name)
+        self.text_splitter = TokenTextSplitter(chunk_size=500, chunk_overlap=30)
+        self.api_key = openai_api_key
+        self.llm = ChatOpenAI(openai_api_key=openai_api_key, temperature=0, model_name=default_chuanhu_assistant_model)
+        PROMPT = PromptTemplate(template=SUMMARIZE_PROMPT, input_variables=["text"])
+        self.summarize_chain = load_summarize_chain(self.llm, chain_type="map_reduce", return_intermediate_steps=True, map_prompt=PROMPT, combine_prompt=PROMPT)
+        if "Pro" in self.model_name:
+            self.tools = load_tools(["google-search-results-json", "llm-math", "arxiv", "wikipedia", "wolfram-alpha"], llm=self.llm)
+        else:
+            self.tools = load_tools(["ddg-search", "llm-math", "arxiv", "wikipedia"], llm=self.llm)
+
+        self.tools.append(
+            Tool.from_function(
+                func=self.summary_url,
+                name="Summary Webpage",
+                description="useful when you need to know the overall content of a webpage.",
+                args_schema=WebBrowsingInput
+            )
+        )
+
+        self.tools.append(
+            StructuredTool.from_function(
+                func=self.ask_url,
+                name="Ask Webpage",
+                description="useful when you need to ask detailed questions about a webpage.",
+                args_schema=WebAskingInput
+            )
+        )
+
+    def summary(self, text):
+        texts = Document(page_content=text)
+        texts = self.text_splitter.split_documents([texts])
+        return self.summarize_chain({"input_documents": texts}, return_only_outputs=True)["output_text"]
+
+    def fetch_url_content(self, url):
+        response = requests.get(url)
+        soup = BeautifulSoup(response.text, 'html.parser')
+
+        # 提取所有的文本
+        text = ''.join(s.getText() for s in soup.find_all('p'))
+        logging.info(f"Extracted text from {url}")
+        return text
+
+    def summary_url(self, url):
+        text = self.fetch_url_content(url)
+        text_summary = self.summary(text)
+        url_content = "webpage content summary:\n" + text_summary
+
+        return url_content
+
+    def ask_url(self, url, question):
+        text = self.fetch_url_content(url)
+        texts = Document(page_content=text)
+        texts = self.text_splitter.split_documents([texts])
+        # use embedding
+        embeddings = OpenAIEmbeddings(openai_api_key=self.api_key)
+
+        # create vectorstore
+        db = FAISS.from_documents(texts, embeddings)
+        retriever = db.as_retriever()
+        qa = RetrievalQA.from_chain_type(llm=self.llm, chain_type="stuff", retriever=retriever)
+        return qa.run(f"{question} Reply in 中文")
+
+    def get_answer_at_once(self):
+        question = self.history[-1]["content"]
+        manager = BaseCallbackManager(handlers=[StdOutCallbackHandler()])
+        # llm=ChatOpenAI(temperature=0, model_name="gpt-3.5-turbo")
+        agent = initialize_agent(self.tools, self.llm, agent=AgentType.STRUCTURED_CHAT_ZERO_SHOT_REACT_DESCRIPTION, verbose=True, callback_manager=manager)
+        reply = agent.run(input=f"{question} Reply in 简体中文")
+        return reply, -1

modules/models/base_model.py

@@ -34,6 +34,7 @@ class ModelType(Enum):
     StableLM = 4
     MOSS = 5
     YuanAI = 6
+    ChuanhuAgent = 7
 
     @classmethod
     def get_type(cls, model_name: str):
@@ -53,6 +54,8 @@ class ModelType(Enum):
             model_type = ModelType.MOSS
         elif "yuanai" in model_name_lower:
             model_type = ModelType.YuanAI
+        elif "川虎助理" in model_name_lower:
+            model_type = ModelType.ChuanhuAgent
         else:
             model_type = ModelType.Unknown
         return model_type
@@ -259,7 +262,6 @@ class BaseLLMModel:
         chatbot,
         stream=False,
         use_websearch=False,
-        autogpt_mode=False,
         files=None,
         reply_language="中文",
         should_check_token_count=True,
@@ -366,7 +368,6 @@ class BaseLLMModel:
         chatbot,
         stream=False,
         use_websearch=False,
-        autogpt_mode=False,
         files=None,
         reply_language="中文",
     ):
@@ -386,7 +387,6 @@ class BaseLLMModel:
             chatbot,
             stream=stream,
             use_websearch=use_websearch,
-            autogpt_mode=autogpt_mode,
             files=files,
             reply_language=reply_language,
         )

modules/models/models.py

@@ -557,6 +557,7 @@ def get_model(
         config.local_embedding = True
     # del current_model.model
     model = None
+    chatbot = gr.Chatbot.update(label=model_name)
     try:
         if model_type == ModelType.OpenAI:
             logging.info(f"正在加载OpenAI模型: {model_name}")
@@ -602,10 +603,12 @@ def get_model(
         elif model_type == ModelType.YuanAI:
             from .inspurai import Yuan_Client
             model = Yuan_Client(model_name, api_key=access_key, user_name=user_name, system_prompt=system_prompt)
+        elif model_type == ModelType.ChuanhuAgent:
+            from .ChuanhuAgent import ChuanhuAgent_Client
+            model = ChuanhuAgent_Client(model_name, access_key, user_name=user_name)
         elif model_type == ModelType.Unknown:
             raise ValueError(f"未知模型: {model_name}")
         logging.info(msg)
-        chatbot = gr.Chatbot.update(label=model_name)
     except Exception as e:
         logging.error(e)
         msg = f"{STANDARD_ERROR_MSG}: {e}"

modules/presets.py

@@ -58,9 +58,9 @@ APPEARANCE_SWITCHER = """
 </div>
 """
 
-SUMMARIZE_PROMPT = "你是谁？我们刚才聊了什么？"  # 总结对话时的 prompt
-
 ONLINE_MODELS = [
+    "川虎助理",
+    "川虎助理 Pro",
     "gpt-3.5-turbo",
     "gpt-3.5-turbo-0301",
     "gpt-4",
@@ -164,6 +164,12 @@ Reply in {reply_language}
 If the context isn't useful, return the original answer.
 """
 
+SUMMARIZE_PROMPT = """Write a concise summary of the following:
+
+{text}
+
+CONCISE SUMMARY IN 中文:"""
+
 ALREADY_CONVERTED_MARK = "<!-- ALREADY CONVERTED BY PARSER. -->"
 
 small_and_beautiful_theme = gr.themes.Soft(