Add README with setup and run instructions for LangGraph + Ollama

- Added steps for virtual environment creation and dependency installation
- Included Ollama installation and server startup guide
- Documented model pulls (gemma, mistral, llama3, nomic-embed-text)
- Provided LangGraph run instructions and browser access info
- Added troubleshooting notes for common errors"

Author: mijeong2aeyo

sample/Rag_example/casts/modules/state.py

@@ -1,7 +1,5 @@
 from langgraph.graph import MessagesState
-from typing import List, Dict, Any
 
 class RagState(MessagesState):
-    """RAG 그래프 상태 (messages + 검색결과 포함)."""
-    retrieved_docs: List[Dict[str, Any]] = []   # RAG 검색 결과
-    context: str = ""                           # 컨텍스트 텍스트
+    """RAG 그래프 상태 (messages 리스트 포함)."""
+    pass

sample/Rag_example/casts/workflow.py

@@ -1,83 +1,106 @@
 import os
-from langgraph.graph import StateGraph, END
-from langchain_community.document_loaders import TextLoader
-from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain_community.embeddings import HuggingFaceInferenceAPIEmbeddings
-from langchain_community.vectorstores import Chroma   # ✅ Chroma로 교체
-from langchain_community.llms import HuggingFaceEndpoint
-
-from .modules.state import RagState
-
-
-# --------------------------
-# 1. 문서 로드 + 분할
-# --------------------------
-def _load_docs():
-    if os.path.exists("README.md"):
-        loader = TextLoader("README.md")
-        docs = loader.load()
-    else:
-        raise FileNotFoundError("README.md를 찾지 못했습니다.")
-    splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
-    return splitter.split_documents(docs)
+from pathlib import Path
+from typing import List
 
+from langgraph.graph import StateGraph, START, END
+from langchain_community.document_loaders import TextLoader
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+from langchain_core.vectorstores import InMemoryVectorStore
+from langchain_community.embeddings import OllamaEmbeddings
+from langchain_community.chat_models import ChatOllama
+
+# ❗ 절대 import (relative import 오류 방지)
+from casts.modules.state import RagState
+
+
+# ----------------------------
+# 0) README 탐색 & 로드
+# ----------------------------
+def _find_readme() -> Path:
+    here = Path(__file__).resolve()
+    candidates = [
+        here.parents[1] / "README.md",   # Rag_Example/README.md
+        here.parents[2] / "README.md",   # 상위 프로젝트 루트/README.md
+    ]
+    for p in candidates:
+        if p.exists():
+            return p
+    raise FileNotFoundError("README.md를 찾지 못했습니다. Rag_Example/README.md 를 생성하세요.")
+
+def _load_docs() -> List[str]:
+    path = _find_readme()
+    loader = TextLoader(str(path), encoding="utf-8")
+    docs = loader.load()
+    splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=80)
+    splits = splitter.split_documents(docs)
+    return [d.page_content for d in splits]
+
+
+# ----------------------------
+# 1) 임베딩/리트리버 (Ollama)
+# ----------------------------
+# Ollama 임베딩 서버: nomic-embed-text
+_embeddings = OllamaEmbeddings(model="nomic-embed-text", base_url=os.getenv("OLLAMA_BASE_URL", "http://localhost:11434"))
 
 _docs = _load_docs()
-
-# --------------------------
-# 2. 임베딩 + 벡터DB (Chroma)
-# --------------------------
-_embeddings = HuggingFaceInferenceAPIEmbeddings(
-    api_key=os.getenv("HF_API_KEY"),  # HuggingFace API 키 (환경변수)
-    model_name="sentence-transformers/all-MiniLM-L6-v2",
-)
-
-vectorstore = Chroma.from_texts([d.page_content for d in _docs], _embeddings)
-retriever = vectorstore.as_retriever()
-
-
-# --------------------------
-# 3. LLM (HuggingFace API)
-# --------------------------
-llm = HuggingFaceEndpoint(
-    repo_id="HuggingFaceH4/zephyr-7b-beta",  # 원하는 모델로 교체 가능
-    huggingfacehub_api_token=os.getenv("HF_API_KEY"),
-    temperature=0.7,
-    max_new_tokens=256,
-    return_full_text=False,
+# InMemory 벡터스토어 → 추가 패키지 설치 불필요
+_vectorstore = InMemoryVectorStore.from_texts(_docs, _embeddings)
+_retriever = _vectorstore.as_retriever(search_kwargs={"k": 4})
+
+# ----------------------------
+# 2) LLM (Ollama Chat)
+# ----------------------------
+# 예: mistral / llama3 등. (미리 pull 필요)
+_llm = ChatOllama(
+    model=os.getenv("OLLAMA_MODEL", "mistral"),
+    base_url=os.getenv("OLLAMA_BASE_URL", "http://localhost:11434"),
+    temperature=0.2
 )
 
 
-# --------------------------
-# 4. 노드 정의
-# --------------------------
-def retrieve_node(state: RagState) -> RagState:
-    """사용자 질문 기반으로 문서 검색"""
-    query = state["messages"][-1]["content"] if state["messages"] else ""
-    docs = retriever.get_relevant_documents(query)
-    state["retrieved_docs"] = [d.page_content for d in docs]
-    return state
-
-
-def generate_node(state: RagState) -> RagState:
-    """검색 결과 + LLM 기반 답변 생성"""
-    query = state["messages"][-1]["content"]
-    context = "\n".join(state.get("retrieved_docs", []))
-    prompt = f"사용자 질문: {query}\n참고 문서:\n{context}\n\n답변:"
-    answer = llm.invoke(prompt)
-    state["messages"].append({"role": "assistant", "content": answer})
-    return state
-
-
-# --------------------------
-# 5. 그래프 정의
-# --------------------------
-workflow = StateGraph(RagState)
-workflow.add_node("retrieve", retrieve_node)
-workflow.add_node("generate", generate_node)
-
-workflow.add_edge("retrieve", "generate")
-workflow.set_entry_point("retrieve")
-workflow.set_finish_point("generate")
-
-rag_workflow = workflow.compile()
+# ----------------------------
+# 3) 유틸
+# ----------------------------
+def _last_user_text(state: RagState) -> str:
+    if not state["messages"]:
+        return ""
+    last = state["messages"][-1]
+    # dict or BaseMessage 모두 안전 처리
+    return last["content"] if isinstance(last, dict) else getattr(last, "content", str(last))
+
+
+# ----------------------------
+# 4) 노드
+# ----------------------------
+def retrieve_node(state: RagState):
+    query = _last_user_text(state)
+    docs = _retriever.get_relevant_documents(query)
+    ctx = "\n\n---\n\n".join(d.page_content for d in docs)
+    # 컨텍스트를 system 메시지로 누적
+    return {"messages": [{"role": "system", "content": f"CONTEXT:\n{ctx}" if ctx else "CONTEXT: (no docs)"}]}
+
+def generate_node(state: RagState):
+    # 누적된 system CONTEXT + user 질문을 하나의 프롬프트로 구성
+    full = "\n\n".join(m["content"] if isinstance(m, dict) else getattr(m, "content", str(m)) for m in state["messages"])
+    prompt = (
+        "You are a helpful RAG assistant. Use the CONTEXT to answer the USER question. "
+        "If the answer is not in the context, say you are not sure.\n\n"
+        f"{full}\n\nAnswer in Korean:"
+    )
+    out = _llm.invoke(prompt)
+    text = out.content if hasattr(out, "content") else str(out)
+    return {"messages": [{"role": "assistant", "content": text}]}
+
+
+# ----------------------------
+# 5) 그래프
+# ----------------------------
+def rag_workflow():
+    g = StateGraph(RagState)
+    g.add_node("retrieve", retrieve_node)
+    g.add_node("generate", generate_node)
+
+    g.add_edge(START, "retrieve")
+    g.add_edge("retrieve", "generate")
+    g.add_edge("generate", END)
+    return g.compile()

sample/Rag_example/langgraph.json

@@ -1,11 +1,9 @@
 {
   "dependencies": [
+    "langgraph",
     "langchain",
     "langchain-community",
-    "gpt4all",
-    "langgraph",
-    "sentence-transformers",
-    "torch"
+    "langchain-text-splitters"
   ],
   "graphs": {
     "main": "./casts/workflow.py:rag_workflow"