Merge remote-tracking branch 'origin/develop-ai'

ai/.gitignore

@@ -5,4 +5,5 @@
 *.pyc
 __pycache__
 data/
-cap30.pem
+cap30.pem
+chatbot.pem

ai/deepl_test.py

@@ -0,0 +1,9 @@
+import deepl
+
+auth_key = 'd31a50f4-1d48-4f54-bbad-c6c63cdcffb3:fx'
+translator = deepl.Translator(auth_key)
+
+message = '안녕하세요. 맛있는 점심 드세요~'
+result = translator.translate_text(message, target_lang='KO')
+
+print(result.text)

ai/llm/deepl_translator.py

@@ -0,0 +1,15 @@
+import deepl
+
+class DeeplTranslator:
+    def __init__(self):
+        API_KEY = 'd31a50f4-1d48-4f54-bbad-c6c63cdcffb3:fx'
+        self.translator = deepl.Translator(API_KEY)
+
+    def fuck(self, message):
+        result = self.translator.translate_text(message, 'ko')
+        return result.text
+
+
+if __name__ == '__main__':
+    dl = DeeplTranslator()
+    dl.fuck()

ai/llm/llm_rag.py

@@ -3,8 +3,9 @@
 from langchain_core.runnables import RunnablePassthrough, RunnableLambda
 from langchain_openai import ChatOpenAI
 from tavily import TavilyClient
-from llm.prompt import casual_prompt, is_qna_prompt, combine_result_prompt, score_prompt, translate_prompt
+from llm.prompt import casual_prompt, is_qna_prompt, combine_result_prompt, score_prompt
 from langchain.retrievers.multi_query import MultiQueryRetriever
+import deepl
 import os
 
 class LLM_RAG:
@@ -18,7 +19,9 @@ def __init__(self, trace = False):
         self.is_qna_prompt = is_qna_prompt()
         self.combine_result_prompt = combine_result_prompt()
         self.score_prompt = score_prompt()
-        self.translate_prompt = translate_prompt()
+        self.deepl = deepl.Translator(os.getenv("DEEPL_API_KEY"))
+        self.ko_query = None
+        self.result_lang = None
         self.notice_retriever = None
         self.school_retriever = None
         self.notice_multiquery_retriever = None
@@ -94,41 +97,37 @@ def set_chain(self):
             self.score_route
         )
 
-        self.translate_chain = (
-            self.translate_prompt
-            | self.llm
-            | StrOutputParser()
-        )
 
     def qna_route(self, info):
         if "question" in info["topic"].lower():
-
-            self.result = self.rag_combine_chain.invoke(info["question"])
-            score = self.score_chain.invoke({"question" : self.question, "answer": self.result})
-            self.score_invoke_chain.invoke({"score" : score, "question": self.question})
+            self.result = self.rag_combine_chain.invoke(self.ko_query)
+            score = self.score_chain.invoke({"question" : self.ko_query, "answer": self.result})
+            self.score_invoke_chain.invoke({"score" : score, "question": self.ko_query})
 
         elif "casual" in info["topic"].lower():
-            self.result =  self.casual_answer_chain.invoke(info['question'])
+            self.result =  self.casual_answer_chain.invoke(self.question)
 
         else:
-            self.result = self.rag_combine_chain.invoke(info["question"])
+            self.result = self.rag_combine_chain.invoke(self.question)
 
 
     def score_route(self, info):
         if "good" in info["score"].lower():
+            self.result = self.deepl.translate_text(self.result, target_lang=self.result_lang).text
             return self.result
         else:
             print('-- google search --')
-            content = self.tavily.qna_search(query='국민대학교 ' + self.question)
-            self.result = "답을 찾을 수 없어서 구글에 검색했습니다.\n\n"
-            self.result += self.translate_chain.invoke({'content' : content, 'question':self.question})
-            return self.result
+            content = self.tavily.qna_search(query='국민대학교 ' + self.ko_query)
+            self.result = "I couldn't find the answer, so I searched on Google.\n\n" + content
+            self.result = self.deepl.translate_text(self.result, target_lang=self.result_lang).text
 
     def format_docs(self, docs):
     # 검색한 문서 결과를 하나의 문단으로 합쳐줍니다.
         return "\n\n".join(doc.page_content + '\nmetadata=' + str(doc.metadata) for doc in docs)
 
-    def query(self, question):
+    def query(self, question, result_lang):
         self.question = question
+        self.ko_query = self.deepl.translate_text(self.question, target_lang='ko').text
+        self.result_lang = result_lang
         self.qna_route_chain.invoke(question)
         return self.result

ai/llm/prompt.py

@@ -39,13 +39,4 @@ def score_prompt():
 
     Classification:"""
     )
-    return prompt
-
-def translate_prompt():
-    prompt = PromptTemplate.from_template("""
-    You are a translator with vast knowledge of human languages. Translate the content into the language corresponding to the question. You should only translate and never answer questions.
-        
-    question : {question}
-    content : {content}
-    result :""")
     return prompt

ai/requirements.txt

@@ -6,4 +6,5 @@ langchainhub==0.1.15
 beautifulsoup4==4.12.3
 pypdf==4.1.0
 python-dotenv==1.0.1
-langchain-text-splitters==0.0.1
+langchain-text-splitters==0.0.1
+deepl

ai/run_chatbot.py

@@ -11,6 +11,7 @@
 load_dotenv()
 
 os.environ['OPENAI_API_KEY'] = os.getenv('OPENAI_API_KEY')
+os.environ['DEEPL_API_KEY'] = os.getenv('DEEPL_API_KEY')
 
 # LangSmith 사용시 아래 주석을 해제
 # os.environ['LANGCHAIN_API_KEY'] = os.getenv('LANGCHAIN_API_KEY')
@@ -40,5 +41,5 @@
     if q == str(0):
         break
     print('AI : ', end='')
-    print(llm.query(q))
+    print(llm.query(q, 'ZH'))
     print()

ai/server.py

@@ -11,6 +11,7 @@
 
 class Query(BaseModel):
     query: str
+    target_lang: str
 
 @asynccontextmanager
 async def lifespan(app:FastAPI):
@@ -58,18 +59,14 @@ async def lifespan(app:FastAPI):
 async def initiate():
     return "안녕하세요! 국민대학교 전용 챗봇 KUKU입니다. 국민대학교에 대한 건 모든 질문해주세요!"
 
-@app.post("/query")
+@app.post("/api/chatbot")
 async def query(query: Query):
-    return {'code': '200',
+    return {'success': 'True',
             'message': 'success',
             'response': {
-                'answer': llm.query(query.query)
+                'answer': llm.query(query.query, query.target_lang)
             }}
 
-@app.post("/input")
-async def input(data: UploadFile):
-    vdb.add_content(data.file)
-    return 
 
 
 if __name__ == "__main__":