source venv/bin/activate
pip install langgraph python-dotenv openai
|
LangGraph Demo
1.0 State
State就是AI流转的全局变量
class InputState(TypedDict):
question: str
llm_answer: Optional[str]
class OutputState(TypedDict):
answer: str
class OverallState(InputState, OutputState):
pass
|
2.0 Node
def llm_node(state: InputeState):
msg = [
("system", readMd("system.md")),
("human", state["question"])
]
llm = ChatOpenAI(model="gpt-4o")
resp = llm.invoke(msg)
return {"answer": resp.content}
|
3.0 Graph Compile
builder = StateGraph(OverallState, input=InputState, output=OutputState)
builder.add_node("llm", llm_node)
builder.add_edge(START, "llm")
builder.add_edge("llm", END)
graph = builder.compile()
|
Draw Graph
display(Image(graph.get_graph(xray=True).draw_mermaid_png()))
|
Messages History
class State(TypedDict):
msgs: Annotated[list, operator.add]
|
MessageGraph
使用Reducer追加消息，但是可以对已有消息做更新、合并、删除操作（Context Engine）
class MessageGraph(StateGraph):
def __init__(self) -> None:
super().__init__(Annotated[list[AnyMessage], add_message])
|
builder = MessageGraph()
graph = builder.compile()
msgs2 = [HumanMessage(content="xxx", id=msg1.id)]
add_messages(msgs1, msgs2)
|
Structured Output
class UserInfo(BaseModel):
name: str = Field(description="The name of the user")
structured_llm = llm.with_structured_output(UserInfo)
resp = structured_llm.invoke(msg)
|