103 lines
3.3 KiB
Python
103 lines
3.3 KiB
Python
import os
|
||
import re
|
||
import asyncio
|
||
from typing import TypedDict, Annotated
|
||
|
||
from langchain_openai import ChatOpenAI
|
||
from langgraph.graph import StateGraph, START, END
|
||
from langgraph.graph.message import add_messages
|
||
|
||
# LLM configuration (OpenRouter)
|
||
llm = ChatOpenAI(
|
||
model="openai/gpt-oss-20b:free",
|
||
base_url="https://openrouter.ai/api/v1",
|
||
api_key=os.getenv("OPENAI_API_KEY"),
|
||
temperature=0.0,
|
||
)
|
||
|
||
# ---------- State definition ----------
|
||
class ReflectState(TypedDict):
|
||
question: str
|
||
draft: str
|
||
critique: str
|
||
verdict: str # ok | needs_revision
|
||
round: int
|
||
max_rounds: int
|
||
|
||
# ---------- Node implementations ----------
|
||
async def draft_answer(state: ReflectState) -> dict:
|
||
prompt = (
|
||
f"Write a concise answer (5–10 sentences) to the following question:\n\n"
|
||
f"Question: {state['question']}"
|
||
)
|
||
response = await llm.ainvoke([{"role": "user", "content": prompt}])
|
||
draft = response.content.strip()
|
||
return {"draft": draft, "round": 0}
|
||
|
||
async def reflect(state: ReflectState) -> dict:
|
||
prompt = (
|
||
f"You are a critical reviewer. Evaluate the following draft answer for completeness, specificity, and lack of filler.\n\n"
|
||
f"Draft: {state['draft']}\n\n"
|
||
f"Provide a verdict (ok or needs_revision) and 2–3 bullet points of critique."
|
||
)
|
||
response = await llm.ainvoke([{"role": "user", "content": prompt}])
|
||
text = response.content.strip()
|
||
verdict_match = re.search(r"(ok|needs_revision)", text, re.IGNORECASE)
|
||
verdict = verdict_match.group(1).lower() if verdict_match else "needs_revision"
|
||
return {"critique": text, "verdict": verdict}
|
||
|
||
async def rewrite(state: ReflectState) -> dict:
|
||
prompt = (
|
||
f"Rewrite the draft answer incorporating the following critique. Keep the answer concise (5–10 sentences).\n\n"
|
||
f"Critique: {state['critique']}\n\n"
|
||
f"Original Draft: {state['draft']}"
|
||
)
|
||
response = await llm.ainvoke([{"role": "user", "content": prompt}])
|
||
new_draft = response.content.strip()
|
||
return {"draft": new_draft, "round": state['round'] + 1}
|
||
|
||
# ---------- Graph construction ----------
|
||
builder = StateGraph(ReflectState)
|
||
builder.add_node("draft_answer", draft_answer)
|
||
builder.add_node("reflect", reflect)
|
||
builder.add_node("rewrite", rewrite)
|
||
|
||
builder.set_entry_point("draft_answer")
|
||
builder.add_edge("draft_answer", "reflect")
|
||
builder.add_conditional_edges(
|
||
"reflect",
|
||
lambda x: x["verdict"],
|
||
{
|
||
"ok": END,
|
||
"needs_revision": "rewrite",
|
||
},
|
||
)
|
||
builder.add_edge("rewrite", "reflect")
|
||
|
||
# Limit rounds
|
||
async def limit_rounds(state: ReflectState) -> str:
|
||
if state["round"] >= state["max_rounds"] and state["verdict"] == "needs_revision":
|
||
return END
|
||
return "reflect"
|
||
|
||
builder.add_conditional_edges("rewrite", limit_rounds, {"reflect": "reflect", END: END})
|
||
|
||
graph = builder.compile()
|
||
|
||
# ---------- Demo execution ----------
|
||
async def main():
|
||
question = "Объясни студенту разницу между tool и resource в MCP."
|
||
initial_state: ReflectState = {
|
||
"question": question,
|
||
"draft": "",
|
||
"critique": "",
|
||
"verdict": "",
|
||
"round": 0,
|
||
"max_rounds": 2,
|
||
}
|
||
final_state = await graph.ainvoke(initial_state)
|
||
print("\nFinal Answer:\n", final_state["draft"])
|
||
|
||
if __name__ == "__main__":
|
||
asyncio.run(main())
|