docGPT 進行 embeddings 時，依然需要使用 OpenAIEmbeddings，因此無法只單純依靠 gpt4free

Author: Lin-jun-xiang

.gitignore

@@ -100,7 +100,8 @@ ipython_config.py
 #   However, in case of collaboration, if having platform-specific dependencies or dependencies
 #   having no cross-platform support, pipenv may install dependencies that don't work, or not
 #   install all needed dependencies.
-#Pipfile.lock
+Pipfile.lock
+
 
 # poetry
 #   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.

Pipfile

@@ -0,0 +1,22 @@
+[[source]]
+url = "https://pypi.org/simple"
+verify_ssl = true
+name = "pypi"
+
+[packages]
+g4f = "==0.0.2.2"
+langchain = "==0.0.218"
+openai = "==0.27.8"
+streamlit = "==1.22.0"
+streamlit-chat = "==0.1.1"
+pymupdf = "==1.22.5"
+faiss-cpu = "==1.7.4"
+tiktoken = "==0.4.0"
+tenacity = "==8.1.0"
+google-search-results = "==2.4.2"
+
+[dev-packages]
+
+[requires]
+python_version = "3.10"
+python_full_version = "3.10.11"

agent/agent.py

@@ -16,14 +16,18 @@
 class AgentHelper:
     """Add agent to help docGPT can be perfonm better."""
     def __init__(self) -> None:
-        self.llm = ChatOpenAI(
-            temperature=0.2,
-            max_tokens=6000,
-            model_name='gpt-3.5-turbo-16k'
-        )
+        self._llm = None
         self.agent_ = None
         self.tools = []
 
+    @property
+    def llm(self):
+        return self._llm
+
+    @llm.setter
+    def llm(self, llm) -> None:
+        self._llm = llm
+
     @property
     def get_calculate_chain(self) -> Tool:
         llm_math_chain = LLMMathChain.from_llm(llm=self.llm, verbose=True)

app.py

@@ -8,12 +8,15 @@
 import langchain
 import streamlit as st
 from langchain.cache import InMemoryCache
+from langchain.chat_models import ChatOpenAI
 from streamlit import logger
 from streamlit_chat import message
 
 from agent import AgentHelper
-from docGPT import DocGPT, OpenAiAPI, SerpAPI
+from docGPT import DocGPT, OpenAiAPI, SerpAPI, GPT4Free
 from model import PDFLoader
+import g4f
+from tenacity import retry, stop_after_attempt
 
 langchain.llm_cache = InMemoryCache()
 
@@ -23,6 +26,7 @@
 
 st.session_state.openai_api_key = None
 st.session_state.serpapi_api_key = None
+st.session_state.g4f_provider = None
 app_logger = logger.get_logger(__name__)
 
 
@@ -66,6 +70,15 @@ def load_api_key() -> None:
 
         os.environ['SERPAPI_API_KEY'] = SERPAPI_API_KEY
 
+    with st.sidebar:
+        st.session_state.g4f_provider = st.selectbox(
+            (
+                "#### Select a provider if you want to use free model. "
+                "([details](https://github.com/xtekky/gpt4free#models))"
+            ),
+            (GPT4Free().PROVIDER_MAPPING.keys())
+        )
+
 
 def upload_and_process_pdf():
     upload_file = st.file_uploader('#### Upload a PDF file:', type='pdf')
@@ -103,12 +116,23 @@ def get_response(query: str):
 with doc_container:
     docs = upload_and_process_pdf()
 
-    agent_, docGPT_tool, calculate_tool, search_tool, llm_tool = [None]*5
-    if OpenAiAPI.is_valid():
+    if docs:
+        docGPT = DocGPT(docs=docs)
+        docGPT_tool, calculate_tool, search_tool, llm_tool = [None]*4
         agent_ = AgentHelper()
 
-        if docs:
-            docGPT = DocGPT(docs=docs)
+        if OpenAiAPI.is_valid():
+            # Use openai llm model
+            docGPT.llm = ChatOpenAI(
+                temperature=0.2,
+                max_tokens=6000,
+                model_name='gpt-3.5-turbo-16k'
+            )
+            agent_.llm = ChatOpenAI(
+                temperature=0.2,
+                max_tokens=6000,
+                model_name='gpt-3.5-turbo-16k'
+            )
             docGPT.create_qa_chain(
                 chain_type='refine',
             )
@@ -117,19 +141,35 @@ def get_response(query: str):
             calculate_tool = agent_.get_calculate_chain
             llm_tool = agent_.create_llm_chain()
 
-    if SerpAPI.is_valid():
-        search_tool = agent_.get_searp_chain
+            if SerpAPI.is_valid():
+                search_tool = agent_.get_searp_chain
+        else:
+            # Use gpt4free llm model
+            docGPT.llm = GPT4Free(
+                provider=GPT4Free().PROVIDER_MAPPING[
+                    st.session_state.g4f_provider
+                ]
+            )
+            agent_.llm = GPT4Free(
+                provider=GPT4Free().PROVIDER_MAPPING[
+                    st.session_state.g4f_provider
+                ]
+            )
+            docGPT.create_qa_chain(
+                chain_type='refine',
+            )
+            docGPT_tool = agent_.create_doc_chat(docGPT)
+        try:
+            tools = [
+                docGPT_tool,
+                search_tool,
+                # llm_tool, # This will cause agent confuse
+                calculate_tool
+            ]
+            agent_.initialize(tools)
+        except Exception as e:
+            app_logger.info(e)
 
-    try:
-        tools = [
-            docGPT_tool,
-            search_tool,
-            # llm_tool, # This will cause agent confuse
-            calculate_tool
-        ]
-        agent_.initialize(tools)
-    except Exception as e:
-        app_logger.info(e)
 
     st.write('---')
 

docGPT/__init__.py

@@ -1,2 +1,2 @@
-from .docGPT import DocGPT
+from .docGPT import DocGPT, GPT4Free
 from .check_api_key import OpenAiAPI, SerpAPI

docGPT/check_api_key.py

@@ -20,7 +20,7 @@ class OpenAiAPI(ApiKey):
     @classmethod
     def is_valid(cls) -> str:
         if not st.session_state['openai_api_key']:
-            st.error('⚠️ :red[You have not pass OpenAI API key.] Necessary Pass')
+            st.error('⚠️ :red[You have not pass OpenAI API key.] Use default key')
             return
 
         openai.api_key = os.getenv('OPENAI_API_KEY')

docGPT/docGPT.py

@@ -1,14 +1,19 @@
 import os
 from abc import ABC, abstractmethod
+from typing import List, Optional
 
+import g4f
 import openai
 from langchain.callbacks import get_openai_callback
+from langchain.callbacks.manager import CallbackManagerForLLMRun
 from langchain.chains import ConversationalRetrievalChain, RetrievalQA
 from langchain.chat_models import ChatOpenAI
 from langchain.embeddings.openai import OpenAIEmbeddings
+from langchain.llms.base import LLM
 from langchain.memory import ConversationBufferMemory
 from langchain.prompts import PromptTemplate
 from langchain.vectorstores import FAISS
+from tenacity import retry, stop_after_attempt
 
 openai.api_key = os.getenv('OPENAI_API_KEY')
 
@@ -80,11 +85,7 @@ class DocGPT:
     def __init__(self, docs):
         self.docs = docs
         self.qa_chain = None
-        self.llm = ChatOpenAI(
-            temperature=0.2,
-            max_tokens=6000,
-            model_name='gpt-3.5-turbo-16k'
-        )
+        self._llm = None
 
         self.prompt_template = """
         Only answer what is asked. Answer step-by-step.
@@ -102,6 +103,14 @@ def __init__(self, docs):
             input_variables=['context', 'question']
         )
 
+    @property
+    def llm(self):
+        return self._llm
+
+    @llm.setter
+    def llm(self, llm) -> None:
+        self._llm = llm
+
     def _helper_prompt(self, chain_type: str) -> None:
         # TODO: Bug helper
         if chain_type == 'refine':
@@ -151,14 +160,14 @@ def create_qa_chain(
             self.qa_chain = RChain(
                 chain_type=chain_type,
                 retriever=retriever,
-                llm=self.llm,
+                llm=self._llm,
                 chain_type_kwargs=chain_type_kwargs
             ).create_qa_chain
         else:
             self.qa_chain = CRChain(
                 chain_type=chain_type,
                 retriever=retriever,
-                llm=self.llm
+                llm=self._llm
             ).create_qa_chain
 
     def run(self, query: str) -> str:
@@ -171,3 +180,53 @@ def run(self, query: str) -> str:
                 response = self.qa_chain({'question': query, 'chat_history': chat_history})
             print(callback)
         return response
+
+
+class GPT4Free(LLM):
+    PROVIDER_MAPPING = {
+        'g4f.Provider.ChatgptAi': g4f.Provider.ChatgptAi,
+        'g4f.Provider.AItianhu': g4f.Provider.AItianhu,
+        'g4f.Provider.Acytoo': g4f.Provider.Acytoo,
+        'g4f.Provider.AiService': g4f.Provider.AiService,
+        'g4f.Provider.Aichat': g4f.Provider.Aichat,
+        'g4f.Provider.Ails': g4f.Provider.Ails,
+        'g4f.Provider.Bard': g4f.Provider.Bard,
+        'g4f.Provider.Bing': g4f.Provider.Bing,
+        'g4f.Provider.ChatgptLogin': g4f.Provider.ChatgptLogin,
+        'g4f.Provider.DeepAi': g4f.Provider.DeepAi,
+        'g4f.Provider.DfeHub': g4f.Provider.DfeHub,
+        'g4f.Provider.EasyChat': g4f.Provider.EasyChat,
+        'g4f.Provider.Forefront': g4f.Provider.Forefront,
+        'g4f.Provider.GetGpt': g4f.Provider.GetGpt,
+        'g4f.Provider.H2o': g4f.Provider.H2o,
+        'g4f.Provider.Liaobots': g4f.Provider.Liaobots,
+        'g4f.Provider.Lockchat': g4f.Provider.Lockchat,
+        'g4f.Provider.Opchatgpts': g4f.Provider.Opchatgpts,
+        'g4f.Provider.Raycast': g4f.Provider.Raycast,
+        'g4f.Provider.Theb': g4f.Provider.Theb,
+        'g4f.Provider.Vercel': g4f.Provider.Vercel,
+        'g4f.Provider.Wewordle': g4f.Provider.Wewordle,
+        'g4f.Provider.You': g4f.Provider.You,
+        'g4f.Provider.Yqcloud': g4f.Provider.Yqcloud,
+    }
+    provider = PROVIDER_MAPPING['g4f.Provider.ChatgptAi']
+
+    @property
+    def _llm_type(self) -> str:
+        return 'gpt4free model'
+
+    # @retry(stop=stop_after_attempt(20))
+    def _call(
+        self,
+        prompt: str,
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[CallbackManagerForLLMRun] = None,
+    ) -> str:
+        try:
+            return g4f.ChatCompletion.create(
+                model="gpt-3.5-turbo",
+                messages=[{"role": "user", "content": prompt}],
+                provider=self.provider
+            )
+        except Exception as e:
+            print(self.provider)

requirements.txt

@@ -1,8 +1,10 @@
+g4f==0.0.2.2
 langchain==0.0.218
 openai==0.27.8
 streamlit==1.22.0
 streamlit_chat==0.1.1
 pymupdf==1.22.5
 faiss-cpu==1.7.4
 tiktoken==0.4.0
+tenacity==8.1.0
 google-search-results==2.4.2