diff --git a/machine-learning/immich_ml/models/ocr/recognition.py b/machine-learning/immich_ml/models/ocr/recognition.py
index 6408e4818f..94f40c9285 100644
--- a/machine-learning/immich_ml/models/ocr/recognition.py
+++ b/machine-learning/immich_ml/models/ocr/recognition.py
@@ -64,6 +64,7 @@ class TextRecognizer(InferenceModel):
                 rec_batch_num=max_batch_size if max_batch_size else 6,
                 rec_img_shape=(3, 48, 320),
                 lang_type=self.language,
+                model_root_dir=self.cache_dir,
             )
         )
         return session
diff --git a/machine-learning/test_main.py b/machine-learning/test_main.py
index b281c0d417..5145be0045 100644
--- a/machine-learning/test_main.py
+++ b/machine-learning/test_main.py
@@ -1028,7 +1028,12 @@ class TestOcr:
         text_recognizer.load()
 
         rapid_recognizer.assert_called_once_with(
-            OcrOptions(session=ort_session.return_value, rec_batch_num=6, rec_img_shape=(3, 48, 320))
+            OcrOptions(
+              session=ort_session.return_value,
+              rec_batch_num=6,
+              rec_img_shape=(3, 48, 320),
+              model_root_dir=text_recognizer.cache_dir,
+            )
         )
 
     def test_set_custom_max_batch_size(self, ort_session: mock.Mock, path: mock.Mock, mocker: MockerFixture) -> None:
@@ -1041,7 +1046,12 @@ class TestOcr:
         text_recognizer.load()
 
         rapid_recognizer.assert_called_once_with(
-            OcrOptions(session=ort_session.return_value, rec_batch_num=4, rec_img_shape=(3, 48, 320))
+            OcrOptions(
+              session=ort_session.return_value,
+              rec_batch_num=4,
+              rec_img_shape=(3, 48, 320),
+              model_root_dir=text_recognizer.cache_dir,
+            )
         )
 
     def test_ignore_other_custom_max_batch_size(
@@ -1056,7 +1066,12 @@ class TestOcr:
         text_recognizer.load()
 
         rapid_recognizer.assert_called_once_with(
-            OcrOptions(session=ort_session.return_value, rec_batch_num=6, rec_img_shape=(3, 48, 320))
+            OcrOptions(
+              session=ort_session.return_value,
+              rec_batch_num=6,
+              rec_img_shape=(3, 48, 320),
+              model_root_dir=text_recognizer.cache_dir,
+            )
         )