diff --git a/docker-compose.yml b/docker-compose.yml
index c0dcf3a..1b320ef 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -46,26 +46,28 @@ services:
       resources:
         limits:
           cpus: '2.0'
-          memory: 8G
+          memory: 6G
         reservations:
           cpus: '1.0'
           memory: 4G
     environment:
       - OLLAMA_HOST=0.0.0.0
       - OLLAMA_ORIGINS=*
-      - OLLAMA_NUM_PARALLEL=2
+      - OLLAMA_NUM_PARALLEL=1
       - OLLAMA_NUM_GPU=0
       - OLLAMA_KEEP_ALIVE=86400
       - OLLAMA_MODEL_TIMEOUT=0
       - OLLAMA_MAX_LOADED_MODELS=1
+      - OLLAMA_FLASH_ATTENTION=0
+      - OLLAMA_LLM_LIBRARY=auto
     healthcheck:
       test: ["CMD", "ollama", "list"]
       interval: 30s
       timeout: 10s
       retries: 5
       start_period: 120s
-    # Предзагружаем модель при запуске контейнера
-    entrypoint: ["/bin/sh", "-c", "ollama serve & sleep 15 && ollama run qwen2.5:7b 'test' && tail -f /dev/null"]
+    # Предзагружаем модель при запуске контейнера с keepalive
+    entrypoint: ["/bin/sh", "-c", "ollama serve & sleep 15 && ollama run --keepalive 24h qwen2.5:7b 'test' && tail -f /dev/null"]
   vector-search:
     build:
       context: ./vector-search
diff --git a/vector-search/app.py b/vector-search/app.py
index 8d52827..eb235dc 100644
--- a/vector-search/app.py
+++ b/vector-search/app.py
@@ -51,7 +51,7 @@ def get_embedding(text: str) -> list:
         resp = requests.post(f"{OLLAMA_BASE_URL}/api/embeddings", json={
             "model": EMBED_MODEL,
             "prompt": text
-        }, timeout=30)
+        }, timeout=300)
         print(f"[DEBUG] Ollama response status: {resp.status_code}")
         
         if not resp.ok: