openviono

export.py

@@ -102,7 +102,27 @@ def convert_openvino():
     ov_text_model = ov.convert_model(traced_text_model, example_input=(inputs.data['input_ids'], inputs.data['attention_mask']))
     ov.save_model(ov_text_model, "openvino/text.xml")
 
-convert_openvino()
+# convert_openvino()
+
+def infer_openvino():
+    import openvino as ov
+    ov_vision_model = ov.Core().read_model("openvino/vision.xml")
+    ov_text_model = ov.Core().read_model("openvino/text.xml")
+
+    compiled_vision_model = ov.Core().compile_model(ov_vision_model, "CPU")
+    compiled_text_model = ov.Core().compile_model(ov_text_model, "CPU")
+
+    vision_predictions = compiled_vision_model(inputs.data['pixel_values'])
+    text_predictions = compiled_text_model((inputs.data['input_ids'], inputs.data['attention_mask']))
+
+    image_embeds = vision_predictions[0]
+    text_embeds = text_predictions[0]
+
+    logits_per_text = text_embeds @ image_embeds.T
+
+    print("similarity:", logits_per_text.item())
+
+infer_openvino()
 
 # convert_coreml()