Trying to load model directly to cpu

app/services/embedding_service.py

@@ -27,7 +27,8 @@ class EmbeddingService:
         # Force the model to load fully into memory without any meta tensors
         torch.cuda.empty_cache() if torch.cuda.is_available() else None
         
-        self.model = SentenceTransformer(model_name, device="cpu")
+        model_args = {'device_map': 'cpu'}
+        self.model = SentenceTransformer(model_name, model_args=model_args, device="cpu")
 
         # Ensure model is fully materialized, not using meta tensors
         for param in self.model.parameters():