9b8059ca38
- reasoner: "이드" 페르소나, 간결+상냥, 불필요한 구조화 금지 - rewriter: 인사/잡담은 그대로 통과 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
97 lines
3.5 KiB
Python
97 lines
3.5 KiB
Python
"""BackendRegistry — 모델 어댑터 관리 + 헬스체크 루프."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import logging
|
|
import time
|
|
|
|
from services.model_adapter import ModelAdapter
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
REWRITER_PROMPT = (
|
|
"너는 질문 재구성 전문가다. "
|
|
"사용자의 질문을 분석하여 의도를 명확히 하고, 구조화된 질문으로 재작성하라. "
|
|
"재구성된 질문만 출력하라. 부연 설명이나 답변은 절대 하지 마라. "
|
|
"인사나 잡담이면 그대로 출력하라."
|
|
)
|
|
|
|
REASONER_PROMPT = (
|
|
"너는 이드, 상냥하고 친근한 AI 어시스턴트야. "
|
|
"간결하고 자연스럽게 대화해. 인사에는 짧게 인사로 답하고, "
|
|
"질문에는 핵심만 명확하게 답해. "
|
|
"불필요한 구조화(번호 매기기, 헤더)는 피하고, 대화하듯 편하게 답변해."
|
|
)
|
|
|
|
|
|
class BackendRegistry:
|
|
def __init__(self) -> None:
|
|
self.rewriter: ModelAdapter | None = None
|
|
self.reasoner: ModelAdapter | None = None
|
|
self._health: dict[str, bool] = {"rewriter": False, "reasoner": False}
|
|
self._latency: dict[str, float] = {"rewriter": 0.0, "reasoner": 0.0}
|
|
self._health_task: asyncio.Task | None = None
|
|
|
|
def init_from_settings(self, settings) -> None:
|
|
self.rewriter = ModelAdapter(
|
|
name="EXAONE",
|
|
base_url=settings.exaone_base_url,
|
|
model=settings.exaone_model,
|
|
system_prompt=REWRITER_PROMPT,
|
|
temperature=settings.exaone_temperature,
|
|
timeout=settings.exaone_timeout,
|
|
)
|
|
self.reasoner = ModelAdapter(
|
|
name="Gemma4",
|
|
base_url=settings.reasoning_base_url,
|
|
model=settings.reasoning_model,
|
|
system_prompt=REASONER_PROMPT,
|
|
temperature=settings.reasoning_temperature,
|
|
timeout=settings.reasoning_timeout,
|
|
)
|
|
|
|
def start_health_loop(self, interval: float = 30.0) -> None:
|
|
self._health_task = asyncio.create_task(self._health_loop(interval))
|
|
|
|
def stop_health_loop(self) -> None:
|
|
if self._health_task and not self._health_task.done():
|
|
self._health_task.cancel()
|
|
|
|
async def _health_loop(self, interval: float) -> None:
|
|
while True:
|
|
await self._check_all()
|
|
await asyncio.sleep(interval)
|
|
|
|
async def _check_all(self) -> None:
|
|
for role, adapter in [("rewriter", self.rewriter), ("reasoner", self.reasoner)]:
|
|
if not adapter:
|
|
continue
|
|
start = time.monotonic()
|
|
healthy = await adapter.health_check()
|
|
elapsed = round((time.monotonic() - start) * 1000, 1)
|
|
prev = self._health[role]
|
|
self._health[role] = healthy
|
|
self._latency[role] = elapsed
|
|
if prev != healthy:
|
|
status = "UP" if healthy else "DOWN"
|
|
logger.warning("%s (%s) → %s (%.0fms)", adapter.name, role, status, elapsed)
|
|
|
|
def is_healthy(self, role: str) -> bool:
|
|
return self._health.get(role, False)
|
|
|
|
def health_summary(self) -> dict:
|
|
result = {}
|
|
for role, adapter in [("rewriter", self.rewriter), ("reasoner", self.reasoner)]:
|
|
if adapter:
|
|
result[role] = {
|
|
"name": adapter.name,
|
|
"model": adapter.model,
|
|
"healthy": self._health[role],
|
|
"latency_ms": self._latency[role],
|
|
}
|
|
return result
|
|
|
|
|
|
backend_registry = BackendRegistry()
|