refactor: Update llm.py and prompt.py files

app.py

@@ -6,7 +6,7 @@ from langchain_core.output_parsers import StrOutputParser
 from langchain_core.runnables import RunnablePassthrough, RunnableLambda
 from langchain_community.document_transformers import LongContextReorder
 
-from libs.config import LLM_MODEL, STREAMING
+from libs.config import STREAMING
 from libs.embeddings import get_embeddings
 from libs.retrievers import load_retrievers
 from libs.llm import get_llm
@@ -22,8 +22,17 @@ embeddings = get_embeddings()
 retriever = load_retrievers(embeddings)
 
 
-def create_rag_chain(chat_history):
-    llm = get_llm(streaming=STREAMING)
+def kiwi_tokenize(text):
+    kiwi = Kiwi()
+    return [token.form for token in kiwi.tokenize(text)]
+
+
+embeddings = get_embeddings()
+retriever = load_retrievers(embeddings)
+
+
+def create_rag_chain(chat_history, model):
+    llm = get_llm(streaming=STREAMING).with_config(configurable={"llm": model})
     prompt = get_prompt(chat_history)
 
     return (
@@ -33,29 +42,64 @@ def create_rag_chain(chat_history):
             "question": RunnablePassthrough(),
         }
         | prompt
-        | llm.with_config(configurable={"llm": LLM_MODEL})
+        | llm
         | StrOutputParser()
     )
 
 
-def respond_stream(message, history):
-    rag_chain = create_rag_chain(history)
-    response = ""
+def respond_stream(message, history, model):
+    rag_chain = create_rag_chain(history, model)
     for chunk in rag_chain.stream(message):
-        response += chunk
-        yield response
+        yield chunk
 
 
-def respond(message, history):
-    rag_chain = create_rag_chain(history)
+def respond(message, history, model):
+    rag_chain = create_rag_chain(history, model)
     return rag_chain.invoke(message)
 
 
-demo = gr.ChatInterface(
-    respond_stream if STREAMING else respond,
-    title="대법원 판례 상담 도우미",
-    description="안녕하세요! 대법원 판례에 관한 질문에 답변해드리는 AI 상담 도우미입니다. 판례 검색, 해석, 적용 등에 대해 궁금하신 점이 있으면 언제든 물어보세요.",
-)
+# 사용 가능한 모델 목록 (key: 모델 식별자, value: 사용자에게 표시할 레이블)
+AVAILABLE_MODELS = {
+    "gpt_3_5_turbo": "GPT-3.5 Turbo",
+    "gpt_4o": "GPT-4o",
+    "claude_3_5_sonnet": "Claude 3.5 Sonnet",
+    "gemini_1_5_flash": "Gemini 1.5 Flash",
+    "llama3_70b": "Llama3 70b",
+}
+
+
+def get_model_key(label):
+    return next(key for key, value in AVAILABLE_MODELS.items() if value == label)
+
+
+def chat_function(message, history, model_label):
+    model_key = get_model_key(model_label)
+    if STREAMING:
+        response = ""
+        for chunk in respond_stream(message, history, model_key):
+            response += chunk
+            yield response
+    else:
+        response = respond(message, history, model_key)
+        yield response
+
+
+with gr.Blocks() as demo:
+    gr.Markdown("# 대법원 판례 상담 도우미")
+    gr.Markdown(
+        "안녕하세요! 대법원 판례에 관한 질문에 답변해드리는 AI 상담 도우미입니다. 판례 검색, 해석, 적용 등에 대해 궁금하신 점이 있으면 언제든 물어보세요."
+    )
+
+    model_dropdown = gr.Dropdown(
+        choices=list(AVAILABLE_MODELS.values()),
+        label="모델 선택",
+        value=list(AVAILABLE_MODELS.values())[1],
+    )
+
+    chatbot = gr.ChatInterface(
+        fn=chat_function,
+        additional_inputs=[model_dropdown],
+    )
 
 if __name__ == "__main__":
     demo.launch()

libs/llm.py

@@ -16,32 +16,32 @@ class StreamCallback(BaseCallbackHandler):
 
 def get_llm(streaming=True):
     return ChatOpenAI(
-        model="gpt-4",
+        model="gpt-4o",
         temperature=0,
         streaming=streaming,
         callbacks=[StreamCallback()],
     ).configurable_alternatives(
         ConfigurableField(id="llm"),
-        default_key="gpt4",
-        claude=ChatAnthropic(
-            model="claude-3-opus-20240229",
+        default_key="gpt_4o",
+        claude_3_5_sonnet=ChatAnthropic(
+            model="claude-3-5-sonnet-20240620",
             temperature=0,
             streaming=streaming,
             callbacks=[StreamCallback()],
         ),
-        gpt3=ChatOpenAI(
+        gpt_3_5_turbo=ChatOpenAI(
             model="gpt-3.5-turbo",
             temperature=0,
             streaming=streaming,
             callbacks=[StreamCallback()],
         ),
-        gemini=GoogleGenerativeAI(
+        gemini_1_5_flash=GoogleGenerativeAI(
             model="gemini-1.5-flash",
             temperature=0,
             streaming=streaming,
             callbacks=[StreamCallback()],
         ),
-        llama3=ChatGroq(
+        llama3_70b=ChatGroq(
             model_name="llama3-70b-8192",
             temperature=0,
             streaming=streaming,