working quantized model

export.py

@@ -180,13 +180,18 @@ def export_openvino_int8():
     vision_dataset = nncf.Dataset(vision_calibration_dataset)
     text_dataset = nncf.Dataset(text_calibration_dataset)
 
-    quantized_vision_model = nncf.quantize(ov_vision_model, vision_dataset, preset=nncf.QuantizationPreset.MIXED)
-    quantized_text_model = nncf.quantize(ov_text_model, text_dataset, preset=nncf.QuantizationPreset.MIXED)
+    quantized_vision_model = nncf.quantize(ov_vision_model, vision_dataset, preset=nncf.QuantizationPreset.MIXED, model_type=nncf.ModelType.TRANSFORMER,
+        # advanced_parameters=nncf.AdvancedQuantizationParameters(disable_bias_correction=True)
+    )
+
+    quantized_text_model = nncf.quantize(ov_text_model, text_dataset, preset=nncf.QuantizationPreset.MIXED, model_type=nncf.ModelType.TRANSFORMER,
+        # advanced_parameters=nncf.AdvancedQuantizationParameters(disable_bias_correction=True)
+    )
 
     ov.save_model(quantized_vision_model, "openvino/vision_int8.xml")
     ov.save_model(quantized_text_model, "openvino/text_int8.xml")
 
-# export_openvino_int8()
+export_openvino_int8()
 
 def infer_openvino_int8():
     import openvino as ov
@@ -246,7 +251,7 @@ def infer_ncnn():
 
     print("similarity:", logits_per_text[0])
 
-infer_ncnn()
+# infer_ncnn()
 
 def infer_torch():
     outputs = ptmodel(**inputs)

openvino/text_int8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad0d394e83a24bc1d2d9415f43a5847ccb804cb44715bbbc45f21b4190673d94
+size 64105628

openvino/vision_int8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:798edc4ba50811ec147985d57d601cf2744e706a6f9dbac0d2d66dd93f159087
+size 88610018