Runnable Version. Experimental changes and new files.

.gitignore

@@ -0,0 +1 @@
+Requirement already satisfied: setuptools in /opt/anaconda3/envs/choozemo-carbon/lib/python3.9/site-packages (70.1.1)

Documents/web_url.csv

@@ -1,5 +1,5 @@
 from dotenv import load_dotenv
-load_dotenv()
+load_dotenv('environment.env')
 
 from langchain_openai import OpenAIEmbeddings
 from langchain_community.embeddings import OllamaEmbeddings
@@ -38,14 +38,15 @@ from ragas.metrics import (
 import pandas as pd
 import os
 import glob
+import openai
 
-from dotenv import load_dotenv
-import os
-load_dotenv()
 URI = os.getenv("SUPABASE_URI")
+openai_api_key = os.getenv("OPENAI_API_KEY")
+openai.api_key = openai_api_key
 
 from RAG_strategy import multi_query, naive_rag
 
+
 def create_retriever(path='Documents', extension="pdf"):
     txt_files = glob.glob(os.path.join(path, f"*.{extension}"))
     
@@ -92,8 +93,10 @@ def create_retriever(path='Documents', extension="pdf"):
 
     # vectorstore
     # vectorstore = Chroma.from_texts(texts=docs, embedding=OpenAIEmbeddings())
-    vectorstore = Chroma.from_documents(documents=docs, embedding=OpenAIEmbeddings())
+    # vectorstore = Chroma.from_documents(documents=docs, embedding=OpenAIEmbeddings(openai_api_key=openai_api_key))
     # vectorstore = Chroma.from_documents(documents=docs, embedding=OllamaEmbeddings(model="llama3", num_gpu=1))
+    vectorstore = Chroma.from_documents(documents=docs, embedding=OllamaEmbeddings(model="gemma2"))
+
     vectorstore.persist()
 
     retriever = vectorstore.as_retriever()

RAG_app.py

@@ -1,3 +1,6 @@
+from dotenv import load_dotenv
+load_dotenv('environment.env')
+
 from fastapi import FastAPI, Request, HTTPException, status, Body
 # from fastapi.templating import Jinja2Templates
 from fastapi.middleware.cors import CORSMiddleware
@@ -26,16 +29,17 @@ from RAG_strategy import multi_query, naive_rag, naive_rag_for_qapairs
 from Indexing_Split import create_retriever as split_retriever
 from Indexing_Split import gen_doc_from_database, gen_doc_from_history
 
-from dotenv import load_dotenv
 import os
 from langchain_community.vectorstores import SupabaseVectorStore
 from langchain_openai import OpenAIEmbeddings
 from supabase.client import Client, create_client
-
-
 from add_vectordb import GetVectorStore
+from langchain_community.cache import RedisSemanticCache  # 更新导入路径
+from langchain_core.prompts import PromptTemplate
+import openai
 
-load_dotenv()
+openai_api_key = os.getenv("OPENAI_API_KEY")
+openai.api_key = openai_api_key
 URI = os.getenv("SUPABASE_URI")
 
 global_retriever = None
@@ -50,12 +54,12 @@ async def lifespan(app: FastAPI):
     # global_retriever = raptor_retriever(path='../Documents', extension="txt")
     # global_retriever = unstructured_retriever(path='../Documents')
 
-    supabase_url = os.environ.get("SUPABASE_URL")
-    supabase_key = os.environ.get("SUPABASE_KEY")
+    supabase_url = os.getenv("SUPABASE_URL")
+    supabase_key = os.getenv("SUPABASE_KEY")
     document_table = "documents"
     supabase: Client = create_client(supabase_url, supabase_key)
 
-    embeddings = OpenAIEmbeddings()
+    embeddings = OpenAIEmbeddings(openai_api_key=openai_api_key)
     vector_store = GetVectorStore(embeddings, supabase, document_table)
     global_retriever = vector_store.as_retriever(search_kwargs={"k": 4})
 
@@ -180,7 +184,15 @@ async def get_history():
     result = loads(result)
     return result.values()
 
+@app.get("/")
+def read_root():
+    return {"message": "Welcome to the SYSTEX API"}
+
+
 if __name__ == "__main__":
-    uvicorn.run("RAG_app:app", host='cmm.ai', port=8081, reload=True, ssl_keyfile="/etc/letsencrypt/live/cmm.ai/privkey.pem", 
-                ssl_certfile="/etc/letsencrypt/live/cmm.ai/fullchain.pem")
+    uvicorn.run("RAG_app:app", host='127.0.0.1', port=8081, reload=True)
+    
+# if __name__ == "__main__":
+#     uvicorn.run("RAG_app:app", host='cmm.ai', port=8081, reload=True, ssl_keyfile="/etc/letsencrypt/live/cmm.ai/privkey.pem", 
+#                 ssl_certfile="/etc/letsencrypt/live/cmm.ai/fullchain.pem")
 

RAG_app_copy.py

@@ -0,0 +1,215 @@
+from dotenv import load_dotenv
+load_dotenv('environment.env')
+
+from fastapi import FastAPI, Request, HTTPException, status, Body
+# from fastapi.templating import Jinja2Templates
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import FileResponse
+from fastapi import Depends
+from contextlib import asynccontextmanager
+from pydantic import BaseModel
+from typing import List, Optional
+import uvicorn
+
+import sqlparse
+from sqlalchemy import create_engine
+import pandas as pd
+#from retrying import retry
+import datetime
+import json
+from json import loads
+import time
+from langchain.callbacks import get_openai_callback
+
+from langchain_community.vectorstores import Chroma
+from langchain_openai import OpenAIEmbeddings
+from RAG_strategy import multi_query, naive_rag, naive_rag_for_qapairs
+from Indexing_Split import create_retriever as split_retriever
+from Indexing_Split import gen_doc_from_database, gen_doc_from_history
+
+import os
+from langchain_community.vectorstores import SupabaseVectorStore
+from langchain_openai import OpenAIEmbeddings
+from supabase.client import Client, create_client
+from add_vectordb import GetVectorStore
+from langchain_community.cache import RedisSemanticCache  # 更新导入路径
+from langchain_core.prompts import PromptTemplate
+import openai
+
+# Get API log
+import logging
+logger = logging.getLogger("uvicorn.error")
+
+openai_api_key = os.getenv("OPENAI_API_KEY")
+URI = os.getenv("SUPABASE_URI")
+openai.api_key = openai_api_key
+
+
+global_retriever = None
+
+# 定義FastAPI的生命週期管理器，在啟動和關閉時執行特定操作
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    global global_retriever
+    global vector_store
+    
+    start = time.time()
+    # global_retriever = split_retriever(path='./Documents', extension="docx")
+    # global_retriever = raptor_retriever(path='../Documents', extension="txt")
+    # global_retriever = unstructured_retriever(path='../Documents')
+
+    supabase_url = os.getenv("SUPABASE_URL")
+    supabase_key = os.getenv("SUPABASE_KEY")
+    URI = os.getenv("SUPABASE_URI")
+    document_table = "documents"
+    supabase: Client = create_client(supabase_url, supabase_key)
+
+    embeddings = OpenAIEmbeddings(openai_api_key=openai_api_key)
+    vector_store = GetVectorStore(embeddings, supabase, document_table)
+    global_retriever = vector_store.as_retriever(search_kwargs={"k": 4})
+
+    print(time.time() - start)
+    yield
+
+# 定義依賴注入函數，用於在請求處理過程中獲取全局變量
+def get_retriever():
+    return global_retriever
+
+
+def get_vector_store():
+    return vector_store
+
+# 創建FastAPI應用實例並配置以及中間件
+app = FastAPI(lifespan=lifespan)
+
+# templates = Jinja2Templates(directory="temp")
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+
+# 定義API路由和處理函數
+# 處理傳入的問題並返回答案
+@app.get("/answer2")
+def multi_query_answer(question, retriever=Depends(get_retriever)):
+    try:
+        start = time.time()
+
+        with get_openai_callback() as cb:
+            # qa_doc = gen_doc_from_database()
+            # qa_history_doc = gen_doc_from_history()
+            # qa_doc.extend(qa_history_doc)
+            # vectorstore = Chroma.from_documents(documents=qa_doc, embedding=OpenAIEmbeddings(), collection_name="qa_pairs")
+            # retriever_qa = vectorstore.as_retriever(search_kwargs={"k": 3})
+            # final_answer, reference_docs = naive_rag_for_qapairs(question, retriever_qa)
+            final_answer = 'False'
+            if final_answer == 'False':
+                final_answer, reference_docs = multi_query(question, retriever, chat_history=[])
+
+        # print(CHAT_HISTORY)
+        
+        # with get_openai_callback() as cb:
+        #     final_answer, reference_docs = multi_query(question, retriever)
+        processing_time = time.time() - start
+        print(processing_time)
+        save_history(question, final_answer, reference_docs, cb, processing_time)
+
+        return {"Answer": final_answer}
+    except Exception as e:
+        logger.error(f"Error in /answer2 endpoint: {e}")
+        raise HTTPException(status_code=500, detail="Internal Server Error")
+
+class ChatHistoryItem(BaseModel):
+    q: str
+    a: str
+
+# 處理帶有歷史聊天紀錄的問題並返回答案
+@app.post("/answer_with_history")
+def multi_query_answer(question: Optional[str] = '', chat_history: List[ChatHistoryItem] = Body(...), retriever=Depends(get_retriever)):
+    start = time.time()
+    
+    chat_history = [(item.q, item.a) for item in chat_history if item.a != ""]
+    print(chat_history)
+
+    # TODO: similarity search
+    
+    with get_openai_callback() as cb:
+        final_answer, reference_docs = multi_query(question, retriever, chat_history)
+    processing_time = time.time() - start
+    print(processing_time)
+    save_history(question, final_answer, reference_docs, cb, processing_time)
+
+    return {"Answer": final_answer}
+
+# 處理帶有聊天歷史紀錄和文件名過濾的問題，並返回答案
+@app.post("/answer_with_history2")
+def multi_query_answer(question: Optional[str] = '', extension: Optional[str] = 'pdf', chat_history: List[ChatHistoryItem] = Body(...), retriever=Depends(get_retriever)):
+    start = time.time()
+
+    retriever = vector_store.as_retriever(search_kwargs={"k": 4,
+                                                         'filter': {'extension':extension}})
+    
+    chat_history = [(item.q, item.a) for item in chat_history if item.a != ""]
+    print(chat_history)
+
+    # TODO: similarity search
+    
+    with get_openai_callback() as cb:
+        final_answer, reference_docs = multi_query(question, retriever, chat_history)
+    processing_time = time.time() - start
+    print(processing_time)
+    save_history(question, final_answer, reference_docs, cb, processing_time)
+
+    return {"Answer": final_answer}
+
+# 保存歷史。將處理結果儲存到數據庫
+def save_history(question, answer, reference, cb, processing_time):
+    # reference = [doc.dict() for doc in reference]
+    record = {
+        'Question': [question],
+        'Answer': [answer],
+        'Total_Tokens': [cb.total_tokens],
+        'Total_Cost': [cb.total_cost],
+        'Processing_time': [processing_time],
+        'Contexts': [str(reference)]
+    }
+    df = pd.DataFrame(record)
+    engine = create_engine(URI)
+    df.to_sql(name='systex_records', con=engine, index=False, if_exists='append')
+
+class history_output(BaseModel):
+    Question: str
+    Answer: str
+    Contexts: str
+    Total_Tokens: int
+    Total_Cost: float
+    Processing_time: float
+    Time: datetime.datetime
+
+# 定義獲取歷史紀錄的路由
+@app.get('/history', response_model=List[history_output])
+async def get_history():
+    engine = create_engine(URI, echo=True)
+
+    df = pd.read_sql_table("systex_records", engine.connect())  
+    df.fillna('', inplace=True)
+    result = df.to_json(orient='index', force_ascii=False)
+    result = loads(result)
+    return result.values()
+
+@app.get("/")
+def read_root():
+    return {"message": "Welcome to the Carbon Chatbot API"}
+
+
+if __name__ == "__main__":
+    uvicorn.run("RAG_app_copy:app", host='127.0.0.1', port=8081, reload=True)
+    
+# if __name__ == "__main__":
+#     uvicorn.run("RAG_app:app", host='cmm.ai', port=8081, reload=True, ssl_keyfile="/etc/letsencrypt/live/cmm.ai/privkey.pem", 
+#                 ssl_certfile="/etc/letsencrypt/live/cmm.ai/fullchain.pem")
+

RAG_strategy.py

@@ -30,20 +30,53 @@ from ragas.metrics import (
     context_precision,
 )
 from typing import List
+import os
 from dotenv import load_dotenv
-load_dotenv()
+load_dotenv('environment.env')
 
 ########################################################################################################################
 ########################################################################################################################
 from langchain.cache import SQLiteCache
-
 from langchain.cache import RedisSemanticCache
 from langchain_openai import OpenAIEmbeddings
-# set_llm_cache(SQLiteCache(database_path=".langchain.db"))
-set_llm_cache(RedisSemanticCache(redis_url="redis://localhost:6380", embedding=OpenAIEmbeddings(), score_threshold=0.0005))
+from langchain.globals import set_llm_cache
+
 ########################################################################################################################
+import requests
+import openai
+openai_api_key = os.getenv("OPENAI_API_KEY")
+openai.api_key = openai_api_key
+URI = os.getenv("SUPABASE_URI")
+
+# 設置緩存，以減少對API的重複請求。使用Redis
+# set_llm_cache(SQLiteCache(database_path=".langchain.db"))
+# set_llm_cache(RedisSemanticCache(redis_url="redis://localhost:6380", embedding=OpenAIEmbeddings(openai_api_key=openai_api_key), score_threshold=0.0005))
+
+# # TAIDE model on Ollama https://ollama.com/jcai/llama3-taide-lx-8b-chat-alpha1
+# def interact_with_model(messages, api_url="http://localhost:11434/v1/chat/completions"):
+#     print("Using model: TAIDE")
+#     response = requests.post(api_url, json={"model": "jcai/llama3-taide-lx-8b-chat-alpha1:Q4_K_M", "messages": messages})
+#     return response.json()["choices"][0]["message"]["content"]
+
+# class CustomTAIDELLM(LLM):
+#     api_url: str = "http://localhost:11434/v1/chat/completions"
+    
+#     def _call(self, prompt: str, stop: Optional[List[str]] = None) -> str:
+#         messages = [{"role": "user", "content": prompt}]
+#         response = requests.post(self.api_url, json={
+#             "model": "taide-local",  # Use your local model name
+#             "messages": messages
+#         })
+#         return response.json()["choices"][0]["message"]["content"]
+    
+#     @property
+#     def _llm_type(self) -> str:
+#         return "custom_taide"
 
+# # Create an instance of the custom LLM
+# taide_llm = CustomTAIDELLM()
 
+# 生成多個不同版本的問題，進行檢索，並返回答案和參考文檔
 def multi_query(question, retriever, chat_history):
 
     def multi_query_chain():
@@ -60,9 +93,18 @@ def multi_query(question, retriever, chat_history):
         Original question: {question}"""
         prompt_perspectives = ChatPromptTemplate.from_template(template)
 
+        messages = [
+            {"role": "system", "content": template},
+            {"role": "user", "content": question},
+        ]
+        # generate_queries = interact_with_model(messages).split("\n")
+
         
-        llm = ChatOpenAI(temperature=0, model="gpt-4-1106-preview")
+        llm = ChatOpenAI(model="gpt-4-1106-preview")
         # llm = ChatOllama(model="llama3", num_gpu=1, temperature=0)
+        # llm = ChatOllama(model="gemma2", temperature=0)
+        # llm = ChatOllama(model=model)
+
 
         generate_queries = (
             prompt_perspectives 
@@ -96,6 +138,7 @@ def multi_query(question, retriever, chat_history):
 
     return answer, docs
 
+# 根據檢索到的文檔和用戶問題生成最後答案
 def multi_query_rag_prompt(retrieval_chain, question):
     # RAG
     template = """Answer the following question based on this context:
@@ -103,16 +146,20 @@ def multi_query_rag_prompt(retrieval_chain, question):
     {context}
 
     Question: {question}
-    Output in user's language. If the question is in zh-tw, then the output will be in zh-tw. \n
+    Output in user's language. If the question is in zh-tw, then the output will be in zh-tw. If the question is in English, then the output will be in English\n
     You should not mention anything about "根據提供的文件內容" or other similar terms.
-    If you don't know the answer, just say that "很抱歉，目前我無法回答您的問題，請將您的詢問發送至 test@systex.com 以便獲得更進一步的幫助，謝謝。"
+    If you don't know the answer, just say that "很抱歉，目前我無法回答您的問題，請將您的詢問發送至 test@email.com 以便獲得更進一步的幫助，謝謝。I'm sorry I cannot answer your question. Please send your question to test@email.com for further assistance. Thank you."
     """
 
     prompt = ChatPromptTemplate.from_template(template)
+    context = retrieval_chain.invoke({"question": question})  # Ensure this returns the context
+
 
     # llm = ChatOpenAI(temperature=0)
-    llm = ChatOpenAI(temperature=0, model="gpt-4-1106-preview")
+    llm = ChatOpenAI(model="gpt-4-1106-preview")
     # llm = ChatOllama(model="llama3", num_gpu=1, temperature=0)
+    # llm = ChatOllama(model="gemma2", temperature=0)
+
 
     final_rag_chain = (
         {"context": retrieval_chain, 
@@ -121,8 +168,13 @@ def multi_query_rag_prompt(retrieval_chain, question):
         | llm
         | StrOutputParser()
     )
-
-    # answer = final_rag_chain.invoke({"question":question})
+    messages = [
+        {"role": "system", "content": template},
+        {"role": "user", "content": question},
+        {"role": "assistant", "content": context}
+    ]
+    # answer = interact_with_model(messages)
+    answer = final_rag_chain.invoke({"question":question})
 
     answer = ""
     for text in final_rag_chain.stream({"question":question}):
@@ -133,6 +185,7 @@ def multi_query_rag_prompt(retrieval_chain, question):
     return answer
 ########################################################################################################################
 
+# 將聊天紀錄個跟進問題轉化為獨立問題
 def get_search_query():
     # Condense a chat history and follow-up question into a standalone question
     # 
@@ -185,7 +238,7 @@ def get_search_query():
                 chat_history=lambda x: _format_chat_history(x["chat_history"])
             )
             | CONDENSE_QUESTION_PROMPT
-            | ChatOpenAI(temperature=0)
+            | ChatOpenAI()
             | StrOutputParser(),
         ),
         # Else, we have no chat history, so just pass through the question
@@ -194,6 +247,7 @@ def get_search_query():
 
     return _search_query
 ########################################################################################################################
+# 檢索文檔並生成答案
 def naive_rag(question, retriever):
     #### RETRIEVAL and GENERATION ####
 
@@ -201,7 +255,7 @@ def naive_rag(question, retriever):
     prompt = hub.pull("rlm/rag-prompt")
 
     # LLM
-    llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)
+    llm = ChatOpenAI(model_name="gpt-3.5-turbo")
 
     # Post-processing
     def format_docs(docs):
@@ -222,6 +276,7 @@ def naive_rag(question, retriever):
 
     return answer, reference
 ################################################################################################
+# 處理question-answer pairs的檢索和生成答案
 def naive_rag_for_qapairs(question, retriever):
     #### RETRIEVAL and GENERATION ####
 
@@ -242,8 +297,10 @@ def naive_rag_for_qapairs(question, retriever):
     prompt = PromptTemplate.from_template(template)
 
     # LLM
-    llm = ChatOpenAI(model_name="gpt-4-0125-preview", temperature=0)
+    llm = ChatOpenAI(model_name="gpt-4-0125-preview")
     # llm = ChatOllama(model="llama3", num_gpu=1, temperature=0)
+    # llm = ChatOllama(model="gemma2", num_gpu=1, temperature=0)
+
 
     # Post-processing
     def format_docs(docs):
@@ -285,4 +342,7 @@ def rag_score(question, ground_truth, answer, reference_docs):
         ],
     )
 
-    return result
+    result_df = result.to_pandas()
+    print(result_df.head())
+    result_df.to_csv('ragas_rag.csv')
+    return result

add_vectordb.py

@@ -1,5 +1,5 @@
 from dotenv import load_dotenv
-load_dotenv()
+load_dotenv('environment.env')
 
 from langchain_openai import OpenAIEmbeddings
 from langchain_community.vectorstores import Chroma
@@ -10,6 +10,7 @@ from langchain_community.document_loaders import Docx2txtLoader
 
 import os
 import glob
+import openai
 
 from langchain_community.vectorstores import SupabaseVectorStore
 from langchain_openai import OpenAIEmbeddings
@@ -156,16 +157,18 @@ if __name__ == "__main__":
     load_dotenv()
     supabase_url = os.environ.get("SUPABASE_URL")
     supabase_key = os.environ.get("SUPABASE_KEY")
+    openai_api_key = os.getenv("OPENAI_API_KEY")
+    openai.api_key = openai_api_key
     document_table = "documents"
     supabase: Client = create_client(supabase_url, supabase_key)
 
-    embeddings = OpenAIEmbeddings()
+    embeddings = OpenAIEmbeddings(openai_api_key=openai_api_key)
 
     # get vector store
     vector_store = GetVectorStore(embeddings, supabase, document_table)
 
     # update data (old + new / all new / all old)
-    path = "/home/mia/systex/Documents"
+    path = "/Documents"
     extension = "pdf"
     # file = None
 

docker-compose.yml

@@ -0,0 +1,20 @@
+version: '3'
+services:
+  ollama:
+    image: ollama/ollama
+    volumes:
+      - ollama:/root/.ollama
+      - /Users/sherry/Documents/_Personal/ChoozeMo/notebooks/carbon/llm/ollama:/models
+    ports:
+      - "11434:11434"
+    mem_limit: 16g
+    cpus: 6
+    command: sh -c "ollama create taide-local -f /models/ide-7b-a.2-q4_k_m.gguf && ollama run taide-local"
+
+  redis:
+    image: redis:alpine
+    ports:
+      - "6379:6379"
+
+volumes:
+  ollama:

environment.env

@@ -0,0 +1,7 @@
+# Choozemo
+SUPABASE_URI = "postgresql://postgres:chuz8310xsystex@db.ptt.cx:5432/postgres"
+SUPABASE_URL = "http://db.ptt.cx:8000/"
+SUPABASE_KEY = "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyAgCiAgICAicm9sZSI6ICJzZXJ2aWNlX3JvbGUiLAogICAgImlzcyI6ICJzdXBhYmFzZS1kZW1vIiwKICAgICJpYXQiOiAxNjQxNzY5MjAwLAogICAgImV4cCI6IDE3OTk1MzU2MDAKfQ.DaYlNEoUrrEn2Ig7tqibS-PHK5vgusbcbo7X36XVt4Q"
+OPENAI_API_KEY = "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"
+HF_API_KEY_7B = "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"
+HF_API_KEY_7B4BIT = "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"

main.log

@@ -0,0 +1,8 @@
+[1719898830] Log start
+[1719898830] Cmd: /usr/local/Cellar/llama.cpp/3274/bin/llama-cli --hf-repo taide/TAIDE-LX-7B-Chat-4bit --hf-file /Users/sherryliu/Documents/_Personal/choozemo/notebooks/carbon/taide-7b-a.2-q4_k_m.gguf -p "I believe the meaning of life is" -n 128
+[1719898830] main: build = 3274 (49122a87)
+[1719898830] main: built with Apple clang version 15.0.0 (clang-1500.3.9.4) for x86_64-apple-darwin23.4.0
+[1719898830] main: seed  = 1719898830
+[1719898830] main: llama backend init
+[1719898830] main: load the model and apply lora adapter, if any
+[1719898830] main: error: unable to load model

out.gv

@@ -0,0 +1,97 @@
+digraph G {
+concentrate=true;
+splines="ortho";
+rankdir="LR";
+subgraph legend{
+    rank = min;
+    label = "legend";
+    Legend [shape=none, margin=0, label = <
+        <table cellspacing="0" cellpadding="0" border="1"><tr><td>Code2flow Legend</td></tr><tr><td>
+        <table cellspacing="0">
+        <tr><td>Regular function</td><td width="50px" bgcolor='#cccccc'></td></tr>
+        <tr><td>Trunk function (nothing calls this)</td><td bgcolor='#966F33'></td></tr>
+        <tr><td>Leaf function (this calls nothing else)</td><td bgcolor='#6db33f'></td></tr>
+        <tr><td>Function call</td><td><font color='black'>&#8594;</font></td></tr>
+        </table></td></tr></table>
+        >];
+}node_7d4b4651 [label="47: lifespan()" name="RAG_app::lifespan" shape="rect" style="rounded,filled" fillcolor="#966F33" ];
+node_4fbd994f [label="116: multi_query_answer()" name="RAG_app::multi_query_answer" shape="rect" style="rounded,filled" fillcolor="#966F33" ];
+node_4c386252 [label="87: multi_query_answer()" name="RAG_app::multi_query_answer" shape="rect" style="rounded,filled" fillcolor="#966F33" ];
+node_9bfa629c [label="134: multi_query_answer()" name="RAG_app::multi_query_answer" shape="rect" style="rounded,filled" fillcolor="#966F33" ];
+node_71a3aeb0 [label="153: save_history()" name="RAG_app::save_history" shape="rect" style="rounded,filled" fillcolor="#6db33f" ];
+node_649e182e [label="147: get_search_query()" name="RAG_strategy::get_search_query" shape="rect" style="rounded,filled" fillcolor="#6db33f" ];
+node_7c28ff05 [label="58: multi_query()" name="RAG_strategy::multi_query" shape="rect" style="rounded,filled" fillcolor="#cccccc" ];
+node_fdd77cdf [label="110: multi_query_rag_prompt()" name="RAG_strategy::multi_query_rag_prompt" shape="rect" style="rounded,filled" fillcolor="#6db33f" ];
+node_4413d1fa [label="0: (global)()" name="add_vectordb::(global)" shape="rect" style="rounded,filled" fillcolor="#966F33" ];
+node_d7743d95 [label="132: __init__()" name="add_vectordb::GetVectorStore.__init__" shape="rect" style="rounded,filled" fillcolor="#cccccc" ];
+node_65de84e4 [label="141: delete()" name="add_vectordb::GetVectorStore.delete" shape="rect" style="rounded,filled" fillcolor="#cccccc" ];
+node_5b1175ed [label="135: insert()" name="add_vectordb::GetVectorStore.insert" shape="rect" style="rounded,filled" fillcolor="#6db33f" ];
+node_e96a8b7c [label="145: update()" name="add_vectordb::GetVectorStore.update" shape="rect" style="rounded,filled" fillcolor="#cccccc" ];
+node_6dc6ba18 [label="125: check_existed_data()" name="add_vectordb::check_existed_data" shape="rect" style="rounded,filled" fillcolor="#6db33f" ];
+node_03dc279e [label="71: create_ids()" name="add_vectordb::create_ids" shape="rect" style="rounded,filled" fillcolor="#6db33f" ];
+node_ba1cc301 [label="20: get_data_list()" name="add_vectordb::get_data_list" shape="rect" style="rounded,filled" fillcolor="#cccccc" ];
+node_76ea7487 [label="104: get_document()" name="add_vectordb::get_document" shape="rect" style="rounded,filled" fillcolor="#cccccc" ];
+node_b5a8f76f [label="35: read_and_split_files()" name="add_vectordb::read_and_split_files" shape="rect" style="rounded,filled" fillcolor="#cccccc" ];
+node_53c93e6f [label="0: (global)()" name="indexing_Split::(global)" shape="rect" style="rounded,filled" fillcolor="#966F33" ];
+node_40bbb064 [label="50: create_retriever()" name="indexing_Split::create_retriever" shape="rect" style="rounded,filled" fillcolor="#cccccc" ];
+node_fa144567 [label="117: gen_doc_from_history()" name="indexing_Split::gen_doc_from_history" shape="rect" style="rounded,filled" fillcolor="#6db33f" ];
+node_7d4b4651 -> node_d7743d95 [color="#E69F00" penwidth="2"];
+node_4c386252 -> node_71a3aeb0 [color="#56B4E9" penwidth="2"];
+node_4c386252 -> node_7c28ff05 [color="#56B4E9" penwidth="2"];
+node_4fbd994f -> node_71a3aeb0 [color="#CC79A7" penwidth="2"];
+node_4fbd994f -> node_7c28ff05 [color="#CC79A7" penwidth="2"];
+node_9bfa629c -> node_71a3aeb0 [color="#F0E442" penwidth="2"];
+node_9bfa629c -> node_7c28ff05 [color="#F0E442" penwidth="2"];
+node_7c28ff05 -> node_649e182e [color="#0072B2" penwidth="2"];
+node_7c28ff05 -> node_fdd77cdf [color="#0072B2" penwidth="2"];
+node_4413d1fa -> node_d7743d95 [color="#56B4E9" penwidth="2"];
+node_4413d1fa -> node_e96a8b7c [color="#56B4E9" penwidth="2"];
+node_4413d1fa -> node_76ea7487 [color="#56B4E9" penwidth="2"];
+node_d7743d95 -> node_d7743d95 [color="#0072B2" penwidth="2"];
+node_65de84e4 -> node_65de84e4 [color="#F0E442" penwidth="2"];
+node_e96a8b7c -> node_65de84e4 [color="#F0E442" penwidth="2"];
+node_e96a8b7c -> node_5b1175ed [color="#F0E442" penwidth="2"];
+node_ba1cc301 -> node_6dc6ba18 [color="#E69F00" penwidth="2"];
+node_76ea7487 -> node_03dc279e [color="#CC79A7" penwidth="2"];
+node_76ea7487 -> node_b5a8f76f [color="#CC79A7" penwidth="2"];
+node_b5a8f76f -> node_ba1cc301 [color="#CC79A7" penwidth="2"];
+node_53c93e6f -> node_7c28ff05 [color="#CC79A7" penwidth="2"];
+node_53c93e6f -> node_7c28ff05 [color="#CC79A7" penwidth="2"];
+node_53c93e6f -> node_40bbb064 [color="#CC79A7" penwidth="2"];
+node_40bbb064 -> node_fa144567 [color="#F0E442" penwidth="2"];
+subgraph cluster_a5d3d6e2 {
+    node_7d4b4651 node_4c386252 node_4fbd994f node_9bfa629c node_71a3aeb0;
+    label="File: RAG_app";
+    name="RAG_app";
+    style="filled";
+    graph[style=dotted];
+};
+subgraph cluster_d5bd3b3c {
+    node_7c28ff05 node_fdd77cdf node_649e182e;
+    label="File: RAG_strategy";
+    name="RAG_strategy";
+    style="filled";
+    graph[style=dotted];
+};
+subgraph cluster_3a83d666 {
+    node_ba1cc301 node_b5a8f76f node_03dc279e node_76ea7487 node_6dc6ba18 node_4413d1fa;
+    label="File: add_vectordb";
+    name="add_vectordb";
+    style="filled";
+    graph[style=dotted];
+    subgraph cluster_622c6110 {
+        node_d7743d95 node_5b1175ed node_65de84e4 node_e96a8b7c;
+        label="Class: GetVectorStore";
+        name="GetVectorStore";
+        style="filled";
+        graph[style=dotted];
+    };
+};
+subgraph cluster_87c7db5c {
+    node_40bbb064 node_fa144567 node_53c93e6f;
+    label="File: indexing_Split";
+    name="indexing_Split";
+    style="filled";
+    graph[style=dotted];
+};
+}

ragas_data_generation.py

@@ -0,0 +1,29 @@
+from dotenv import load_dotenv
+load_dotenv('environment.env')
+
+
+from ragas.testset.generator import TestsetGenerator
+from ragas.testset.evolutions import simple, reasoning, multi_context 
+from langchain_openai import ChatOpenAi, OpenAIEmbeddings
+from langchain_community.document_loaders import DirectoryLoader
+from langchain_community.document_loaders import PyPDFLoader
+
+loader = DirectoryLoader("Documents")
+for file in 
+documents = loader.load()
+
+
+for document in documents:
+    document.metadata['filename'] = document.metadata['source']
+
+generator_llm = ChatOpenAi(model = "gpt-3.5-turbo-16k")
+critic_llm = ChatOpenAI(model="gpt-4")
+embeddings = OpenAIEmbeddings()
+
+generator = TestGenerator.from_langchain(
+    generator_llm,
+    critic_llm,
+    embeddings
+)
+# Generate testset
+testset = generator.generate_with_langchain_docs(documents, test_size=10, distributions={simple: 0.5, reasoning: 0.25, multi_context: 0.25})

requirements.txt

@@ -140,3 +140,13 @@ wrapt==1.16.0
 xxhash==3.4.1
 yarl==1.9.4
 zipp==3.18.1
+python-dotenv
+fastapi
+sqlparse
+sqlalchemy
+pandas
+langchain
+datasets
+ragas
+supabase
+psycopg2-binary

requirements_semantic_search.txt

@@ -0,0 +1,83 @@
+#!/bin/bash
+
+# Set environment variables
+export OPENAI_API_KEY="sk-proj-kGQPjKXup6g5QmjWvN3GT3BlbkFJDOYyhv8auoHBDIznmbgj"
+export SUPABASE_URI=postgresql://postgres:chuz8310xsystex@db.ptt.cx:5432/postgres
+export SUPABASE_URL="http://db.ptt.cx:8000/"
+export SUPABASE_KEY="eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyAgCiAgICAicm9sZSI6ICJzZXJ2aWNlX3JvbGUiLAogICAgImlzcyI6ICJzdXBhYmFzZS1kZW1vIiwKICAgICJpYXQiOiAxNjQxNzY5MjAwLAogICAgImV4cCI6IDE3OTk1MzU2MDAKfQ.DaYlNEoUrrEn2Ig7tqibS-PHK5vgusbcbo7X36XVt4Q"
+
+
+Function to check if Docker is running
+start_docker() {
+    if ! docker info > /dev/null 2>&1; then
+        echo "Starting Docker Desktop..."
+        open --background -a Docker
+        # Wait until Docker daemon is running
+        while ! docker info > /dev/null 2>&1; do
+            echo "Waiting for Docker to start..."
+            sleep 1
+        done
+        echo "Docker started."
+    else
+        echo "Docker is already running."
+    fi
+}
+
+#########################################
+# # Start Docker if it's not already running
+# if ! systemctl is-active --quiet docker; then
+#     echo "Starting Docker..."
+#     sudo systemctl start docker
+#     sleep 5  # Wait for Docker to fully start
+# fi
+
+# # Get the script directory
+# script_dir=$(dirname "$0")
+
+# # Change to the directory containing the docker-compose.yml file
+# cd "$script_dir"
+
+# # Start the services defined in docker-compose.yml
+# echo "Starting services with Docker Compose..."
+# docker-compose up -d
+
+# # Wait for services to be fully up
+# echo "Waiting for services to start..."
+# sleep 20  # Adjust this time as needed
+
+# # Change to the directory containing your Python script
+# cd "$script_dir/systex-RAG-sherry"
+
+# Run your Python script
+# echo "Running RAG application..."
+# python RAG_app_copy.py
+
+##########################################
+
+# Check and manage Redis Stack container
+if [ "$(docker ps -a -q -f name=redis-stack)" ]; then
+    if [ ! "$(docker ps -q -f name=redis-stack)" ]; then
+        echo "Starting existing Redis Stack container..."
+        docker start redis-stack
+    else
+        echo "Redis Stack container is already running."
+    fi
+else
+    echo "Creating and starting Redis Stack container..."
+    docker run -d --name redis-stack -p 6380:6379 redis/redis-stack:latest
+fi
+
+# Check if Redis container started successfully
+if [ ! "$(docker ps -q -f name=redis-stack)" ]; then
+    echo "Redis Stack container failed to start. Exiting script."
+    exit 1
+fi
+
+echo "Running RAG application..."
+python RAG_app_copy.py
+
+# 使脚本文件可执行：
+# chmod +x run.sh
+
+# /answer2?question=解釋碳排放法規
+

test_connection.py

@@ -0,0 +1,102 @@
+# import os
+# import sys
+
+# from supabase import create_client, Client
+
+# # # Load environment variables
+# from dotenv import load_dotenv
+# load_dotenv('environment.env')
+
+# # Get Supabase configuration from environment variables
+# SUPABASE_URL = os.getenv("SUPABASE_URL")
+# SUPABASE_KEY = os.getenv("SUPABASE_KEY")
+# SUPABASE_URI = os.getenv("SUPABASE_URI")
+# OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
+
+# # Check if environment variables are successfully loaded
+# if not SUPABASE_URL or not SUPABASE_KEY or not OPENAI_API_KEY or not SUPABASE_URI:
+#     print("Please ensure SUPABASE_URL, SUPABASE_KEY, and OPENAI_API_KEY are correctly set in the .env file.")
+#     sys.exit(1)
+# else:
+#     print("Connection successful.")
+#     try:
+#         supabase: Client = create_client(SUPABASE_URL, SUPABASE_KEY)
+#         print("Client created successfully.")
+#     except Exception as e:
+#         print("Client creation failed:", e)
+#         sys.exit(1)
+
+# # List all table names
+# try:
+#     response = supabase.table('information_schema.tables').select('table_name').eq('table_schema', 'public').execute()
+#     table_names = [table['table_name'] for table in response.data]
+#     print("All table names:")
+#     for name in table_names:
+#         print(name)
+# except Exception as e:
+#     print("Connection failed:", e)
+#     sys.exit(1)
+
+
+# ### Test hugging face tokens for the TAIDE local model. ######################################################
+# from transformers import AutoTokenizer, AutoModelForCausalLM
+
+# token = os.getenv("HF_API_KEY_7B4BIT")
+
+# # Check if the token is loaded correctly
+# if token is None:
+#     raise ValueError("Hugging Face API token is not set. Please check your environment.env file.")
+
+# # Load the tokenizer and model with the token
+# try:
+#     tokenizer = AutoTokenizer.from_pretrained("../TAIDE-LX-7B-Chat-4bit", token=token)  
+#     model = AutoModelForCausalLM.from_pretrained("../TAIDE-LX-7B-Chat-4bit", token=token)
+    
+#     # Verify the model and tokenizer
+#     print(f"Loaded tokenizer: {tokenizer.name_or_path}")
+#     print(f"Loaded model: {model.name_or_path}")
+
+#     # Optional: Print model and tokenizer configuration for more details
+#     print(f"Model configuration: {model.config}")
+#     print(f"Tokenizer configuration: {tokenizer}")
+
+# except Exception as e:
+#     print(f"Error loading model or tokenizer: {e}")
+
+#################################################################################################################
+# import torch
+# from transformers import AutoModelForCausalLM, AutoTokenizer
+# from huggingface_hub import hf_hub_download
+# from llama_cpp import Llama
+
+# ## Download the GGUF model
+# model_name = "TheBloke/Mixtral-8x7B-Instruct-v0.1-GGUF"
+# model_file = "mixtral-8x7b-instruct-v0.1.Q4_K_M.gguf" # this is the specific model file we'll use in this example. It's a 4-bit quant, but other levels of quantization are available in the model repo if preferred
+# model_path = hf_hub_download(model_name, filename=model_file)
+
+
+
+
+# import requests
+
+# def generate_response(input_text, max_length=512, temperature=0.7):
+#     # URL to interact with the model
+#     url = "http://localhost:11434/v1/chat/completions"  # Adjust based on how Ollama exposes the model
+
+#     # Payload to send to the model
+#     payload = {
+#         "input": input_text,
+#         "parameters": {
+#             "max_length": max_length,
+#             "temperature": temperature
+#         }
+#     }
+
+#     # Make a request to the model
+#     response = requests.post(url, json=payload)
+#     return response.json()["output"]
+
+# if __name__ == "__main__":
+#     input_text = "I believe the meaning of life is"
+#     response = generate_response(input_text, max_length=128, temperature=0.5)
+#     print(f"Model: {response}")