refactor: AI 서비스 조립컴 Ollama 단독 운영으로 전환

- Ollama 메인 → MLX fallback 순서로 변경 (기존 MLX 우선 제거)
- OLLAMA_BASE_URL을 gpu.hyungi.net으로 변경 (Docker 네트워크 호환)
- OLLAMA_TEXT_MODEL을 qwen3:8b → qwen3.5:9b-q8_0으로 업데이트
- health 엔드포인트: model 필드 직접 반환, 이중 중첩 해소
- health 체크 타임아웃 120초 → 5초로 단축
- Ollama API 호출에 think: false 추가 (thinking 토큰 방지)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ai-service/config.py

@@ -2,8 +2,8 @@ from pydantic_settings import BaseSettings
 
 
 class Settings(BaseSettings):
-    OLLAMA_BASE_URL: str = "http://100.111.160.84:11434"
+    OLLAMA_BASE_URL: str = "https://gpu.hyungi.net"
-    OLLAMA_TEXT_MODEL: str = "qwen3:8b"
+    OLLAMA_TEXT_MODEL: str = "qwen3.5:9b-q8_0"
     OLLAMA_EMBED_MODEL: str = "bge-m3"
     OLLAMA_TIMEOUT: int = 120
 

ai-service/routers/health.py

@@ -7,12 +7,24 @@ router = APIRouter(tags=["health"])
 
 @router.get("/health")
 async def health_check():
-    ollama_status = await ollama_client.check_health()
+    backends = await ollama_client.check_health()
+    stats = vector_store.stats()
+
+    # 메인 텍스트 모델명 결정 (Ollama 메인, MLX fallback)
+    model_name = None
+    ollama_models = backends.get("ollama", {}).get("models", [])
+    if ollama_models:
+        model_name = ollama_models[0]
+    if not model_name and backends.get("mlx", {}).get("status") == "connected":
+        model_name = backends["mlx"].get("model")
+
     return {
         "status": "ok",
         "service": "tk-ai-service",
-        "ollama": ollama_status,
+        "model": model_name,
-        "embeddings": vector_store.stats(),
+        "ollama": backends.get("ollama", {}),
+        "mlx": backends.get("mlx", {}),
+        "embeddings": stats,
     }
 
 

ai-service/services/ollama_client.py

@@ -43,7 +43,21 @@ class OllamaClient:
             messages.append({"role": "system", "content": system})
         messages.append({"role": "user", "content": prompt})
         client = await self._get_client()
+        # 조립컴 Ollama 메인, MLX fallback
         try:
+            response = await client.post(
+                f"{self.base_url}/api/chat",
+                json={
+                    "model": settings.OLLAMA_TEXT_MODEL,
+                    "messages": messages,
+                    "stream": False,
+                    "think": False,
+                    "options": {"temperature": 0.3, "num_predict": 2048},
+                },
+            )
+            response.raise_for_status()
+            return response.json()["message"]["content"]
+        except Exception:
             response = await client.post(
                 f"{settings.MLX_BASE_URL}/chat/completions",
                 json={
@@ -55,31 +69,20 @@ class OllamaClient:
             )
             response.raise_for_status()
             return response.json()["choices"][0]["message"]["content"]
-        except Exception:
-            response = await client.post(
-                f"{self.base_url}/api/chat",
-                json={
-                    "model": settings.OLLAMA_TEXT_MODEL,
-                    "messages": messages,
-                    "stream": False,
-                    "options": {"temperature": 0.3, "num_predict": 2048},
-                },
-            )
-            response.raise_for_status()
-            return response.json()["message"]["content"]
 
     async def check_health(self) -> dict:
         result = {}
+        short_timeout = httpx.Timeout(5.0, connect=3.0)
         try:
-            client = await self._get_client()
+            async with httpx.AsyncClient(timeout=short_timeout) as c:
-            response = await client.get(f"{self.base_url}/api/tags")
+                response = await c.get(f"{self.base_url}/api/tags")
             models = response.json().get("models", [])
             result["ollama"] = {"status": "connected", "models": [m["name"] for m in models]}
         except Exception:
             result["ollama"] = {"status": "disconnected"}
         try:
-            client = await self._get_client()
+            async with httpx.AsyncClient(timeout=short_timeout) as c:
-            response = await client.get(f"{settings.MLX_BASE_URL}/health")
+                response = await c.get(f"{settings.MLX_BASE_URL}/health")
             result["mlx"] = {"status": "connected", "model": settings.MLX_TEXT_MODEL}
         except Exception:
             result["mlx"] = {"status": "disconnected"}

docker-compose.yml

@@ -298,8 +298,8 @@ services:
     ports:
       - "30400:8000"
     environment:
-      - OLLAMA_BASE_URL=${OLLAMA_BASE_URL:-http://100.111.160.84:11434}
+      - OLLAMA_BASE_URL=${OLLAMA_BASE_URL:-https://gpu.hyungi.net}
-      - OLLAMA_TEXT_MODEL=${OLLAMA_TEXT_MODEL:-qwen3:8b}
+      - OLLAMA_TEXT_MODEL=${OLLAMA_TEXT_MODEL:-qwen3.5:9b-q8_0}
       - OLLAMA_EMBED_MODEL=${OLLAMA_EMBED_MODEL:-bge-m3}
       - OLLAMA_TIMEOUT=${OLLAMA_TIMEOUT:-120}
       - MLX_BASE_URL=${MLX_BASE_URL:-https://llm.hyungi.net}