RAG, MMR, 임베딩 심화

ollama/food2.py

@@ -9,11 +9,13 @@ from langchain_core.output_parsers import (
 from pydantic import BaseModel, Field
 from typing import Literal
 from dotenv import load_dotenv
+from langchain_core.prompts import MessagesPlaceholder # 추가 필수
 import os
 import gradio as gr
 
 qwen_llm = ChatOllama(model="qwen3.5:4b")
 exaone_llm = ChatOllama(model="exaone3.5:2.4b")
+gemma_llm = ChatOllama(model="gemma4:e2b")
 
 system_prompt = """\
     당신은 20년 경력의 전문 셰프이자 요리 연구가입니다.
@@ -26,15 +28,23 @@ system_prompt = """\
 template = ChatPromptTemplate.from_messages(
     [
         ("system", system_prompt),
+        MessagesPlaceholder(variable_name="history"), # 대화 이력이 삽입될 위치
         ("human", "{question}"),
     ]
 )
 
-chain = template | exaone_llm | StrOutputParser()
-
+chain = template | gemma_llm | StrOutputParser()
 
 def chat(question, history):
-    response = chain.invoke({"question": question})
+    chat_history = []
+    # history 를 이용한 직접 관리
+    for msg in history:
+        if msg["role"] == "user":
+            chat_history.append(["human", msg["content"]])
+        elif msg["role"] == "assistant":
+            chat_history.append(["ai", msg["content"]])
+
+    response = chain.invoke({"history": chat_history, "question": question})
     return response