diff --git a/src/super_gradients/training/sg_trainer/sg_trainer.py b/src/super_gradients/training/sg_trainer/sg_trainer.py
index 069668b186..cd433eefd6 100755
--- a/src/super_gradients/training/sg_trainer/sg_trainer.py
+++ b/src/super_gradients/training/sg_trainer/sg_trainer.py
@@ -324,12 +324,14 @@ def evaluate_from_recipe(cls, cfg: DictConfig) -> Tuple[nn.Module, Tuple]:
             name=cfg.val_dataloader, dataset_params=cfg.dataset_params.val_dataset_params, dataloader_params=cfg.dataset_params.val_dataloader_params
         )
 
-        if cfg.checkpoint_params.pretrained_weights is None and cfg.checkpoint_params.checkpoint_path is None:
+        if cfg.checkpoint_params.checkpoint_path is None:
             logger.info(
                 "checkpoint_params.checkpoint_path was not provided, " "so the recipe will be evaluated using checkpoints_dir/training_hyperparams.ckpt_name"
             )
-            checkpoints_dir = Path(get_checkpoints_dir_path(experiment_name=cfg.experiment_name, ckpt_root_dir=cfg.ckpt_root_dir))
-            cfg.checkpoint_params.checkpoint_path = str(checkpoints_dir / cfg.training_hyperparams.ckpt_name)
+            checkpoints_dir = get_checkpoints_dir_path(experiment_name=cfg.experiment_name, ckpt_root_dir=cfg.ckpt_root_dir)
+            checkpoint_path = os.path.join(checkpoints_dir, cfg.training_hyperparams.ckpt_name)
+            if os.path.exists(checkpoint_path):
+                cfg.checkpoint_params.checkpoint_path = checkpoint_path
 
         logger.info(f"Evaluating checkpoint: {cfg.checkpoint_params.checkpoint_path}")