diff --git a/machine-learning/app/models/base.py b/machine-learning/app/models/base.py
index 496012270..16adc159a 100644
--- a/machine-learning/app/models/base.py
+++ b/machine-learning/app/models/base.py
@@ -186,7 +186,7 @@ class InferenceModel(ABC):
                 case "CPUExecutionProvider" | "CUDAExecutionProvider":
                     option = {"arena_extend_strategy": "kSameAsRequested"}
                 case "OpenVINOExecutionProvider":
-                    option = {"device_type": "GPU_FP32"}
+                    option = {"device_type": "GPU_FP32", "cache_dir": (self.cache_dir / "openvino").as_posix()}
                 case _:
                     option = {}
             options.append(option)
diff --git a/machine-learning/app/test_main.py b/machine-learning/app/test_main.py
index d79da0cbd..22038a654 100644
--- a/machine-learning/app/test_main.py
+++ b/machine-learning/app/test_main.py
@@ -88,7 +88,7 @@ class TestBase:
         encoder = OpenCLIPEncoder("ViT-B-32__openai", providers=["OpenVINOExecutionProvider", "CPUExecutionProvider"])
 
         assert encoder.provider_options == [
-            {"device_type": "GPU_FP32"},
+            {"device_type": "GPU_FP32", "cache_dir": (encoder.cache_dir / "openvino").as_posix()},
             {"arena_extend_strategy": "kSameAsRequested"},
         ]