Implement model preloading in PoetryGenerationService for faster startup; remove redundant preload method from ModelManager

app/services/poetry_generation.py

@@ -37,7 +37,16 @@ class PoetryGenerationService:
             "emotional": "Create an emotional poem about {}:\n",
             "memory": "Write a memory-focused poem about {}:\n"
         }
-
+    def preload_models(self):
+        """Minimal preload for faster startup"""
+        try:
+            logger.info("Starting minimal model preload...")
+            _ = self.generate_poem("test", max_new_tokens=20)
+            logger.info("Basic preload completed")
+            return True
+        except Exception as e:
+            logger.error(f"Error preloading models: {str(e)}")
+            raise
     @lru_cache(maxsize=100)
     def clean_output(self, text: str) -> str:
         """Optimized text cleaning with caching"""
@@ -195,13 +204,3 @@ class ModelManager:
             torch.cuda.empty_cache()
             gc.collect()
 
-    def preload_models(self):
-        """Minimal preload for faster startup"""
-        try:
-            logger.info("Starting minimal model preload...")
-            _ = self.generate_poem("test", max_new_tokens=20)
-            logger.info("Basic preload completed")
-            return True
-        except Exception as e:
-            logger.error(f"Error preloading models: {str(e)}")
-            raise