migrate to ov.save_model(...)

openvinotoolkit · Oct 25, 2023 · 3056252 · 3056252
1 parent 9cc57c0
commit 3056252
Show file tree

Hide file tree

Showing 6 changed files with 10 additions and 10 deletions.
diff --git a/.../post_training_quantization/openvino/anomaly_stfpm_quantize_with_accuracy_control/main.py b/.../post_training_quantization/openvino/anomaly_stfpm_quantize_with_accuracy_control/main.py
@@ -165,12 +165,12 @@ def transform_fn(data_item):
 # Benchmark performance, calculate compression rate and validate accuracy
 
 fp32_ir_path = f"{ROOT}/stfpm_fp32.xml"
-ov.serialize(ov_model, fp32_ir_path)
+ov.save_model(ov_model, fp32_ir_path, compress_to_fp16=False)
 print(f"[1/7] Save FP32 model: {fp32_ir_path}")
 fp32_size = get_model_size(fp32_ir_path, verbose=True)
 
 int8_ir_path = f"{ROOT}/stfpm_int8.xml"
-ov.serialize(ov_quantized_model, int8_ir_path)
+ov.save_model(ov_quantized_model, int8_ir_path, compress_to_fp16=False)
 print(f"[2/7] Save INT8 model: {int8_ir_path}")
 int8_size = get_model_size(int8_ir_path, verbose=True)
 

diff --git a/examples/post_training_quantization/openvino/mobilenet_v2/main.py b/examples/post_training_quantization/openvino/mobilenet_v2/main.py
@@ -137,12 +137,12 @@ def transform_fn(data_item):
 # Benchmark performance, calculate compression rate and validate accuracy
 
 fp32_ir_path = f"{ROOT}/mobilenet_v2_fp32.xml"
-ov.serialize(ov_model, fp32_ir_path)
+ov.save_model(ov_model, fp32_ir_path, compress_to_fp16=False)
 print(f"[1/7] Save FP32 model: {fp32_ir_path}")
 fp32_model_size = get_model_size(fp32_ir_path, verbose=True)
 
 int8_ir_path = f"{ROOT}/mobilenet_v2_int8.xml"
-ov.serialize(ov_quantized_model, int8_ir_path)
+ov.save_model(ov_quantized_model, int8_ir_path, compress_to_fp16=False)
 print(f"[2/7] Save INT8 model: {int8_ir_path}")
 int8_model_size = get_model_size(int8_ir_path, verbose=True)
 

diff --git a/examples/post_training_quantization/openvino/yolov8/main.py b/examples/post_training_quantization/openvino/yolov8/main.py
@@ -158,7 +158,7 @@ def main():
     # Quantize mode in OpenVINO representation
     quantized_model = quantize(ov_model, data_loader, validator)
     quantized_model_path = Path(f"{ROOT}/{MODEL_NAME}_openvino_model/{MODEL_NAME}_quantized.xml")
-    ov.serialize(quantized_model, str(quantized_model_path))
+    ov.save_model(quantized_model, str(quantized_model_path), compress_to_fp16=False)
 
     # Validate FP32 model
     fp_stats, total_images, total_objects = validate(ov_model, tqdm(data_loader), validator)

diff --git a/examples/post_training_quantization/openvino/yolov8_quantize_with_accuracy_control/main.py b/examples/post_training_quantization/openvino/yolov8_quantize_with_accuracy_control/main.py
@@ -222,7 +222,7 @@ def main():
     quantized_model = quantize_ac(ov_model, data_loader, validator)
 
     quantized_model_path = Path(f"{ROOT}/{MODEL_NAME}_openvino_model/{MODEL_NAME}_quantized.xml")
-    ov.serialize(quantized_model, str(quantized_model_path))
+    ov.save_model(quantized_model, str(quantized_model_path), compress_to_fp16=False)
 
     # Validate FP32 model
     fp_stats, total_images, total_objects = validate(ov_model, tqdm(data_loader), validator)

diff --git a/examples/post_training_quantization/torch/mobilenet_v2/main.py b/examples/post_training_quantization/torch/mobilenet_v2/main.py
@@ -173,12 +173,12 @@ def transform_fn(data_item):
 ov_quantized_model = mo.convert_model(int8_onnx_path)
 
 fp32_ir_path = f"{ROOT}/mobilenet_v2_fp32.xml"
-ov.serialize(ov_model, fp32_ir_path)
+ov.save_model(ov_model, fp32_ir_path, compress_to_fp16=False)
 print(f"[1/7] Save FP32 model: {fp32_ir_path}")
 fp32_model_size = get_model_size(fp32_ir_path, verbose=True)
 
 int8_ir_path = f"{ROOT}/mobilenet_v2_int8.xml"
-ov.serialize(ov_quantized_model, int8_ir_path)
+ov.save_model(ov_quantized_model, int8_ir_path, compress_to_fp16=False)
 print(f"[2/7] Save INT8 model: {int8_ir_path}")
 int8_model_size = get_model_size(int8_ir_path, verbose=True)
 

diff --git a/examples/post_training_quantization/torch/ssd300_vgg16/main.py b/examples/post_training_quantization/torch/ssd300_vgg16/main.py
@@ -163,12 +163,12 @@ def main():
     ov_quantized_model = mo.convert_model(int8_onnx_path)
 
     fp32_ir_path = f"{ROOT}/ssd300_vgg16_fp32.xml"
-    ov.serialize(ov_model, fp32_ir_path)
+    ov.save_model(ov_model, fp32_ir_path, compress_to_fp16=False)
     print(f"[1/7] Save FP32 model: {fp32_ir_path}")
     fp32_model_size = get_model_size(fp32_ir_path, verbose=True)
 
     int8_ir_path = f"{ROOT}/ssd300_vgg16_int8.xml"
-    ov.serialize(ov_quantized_model, int8_ir_path)
+    ov.save_model(ov_quantized_model, int8_ir_path, compress_to_fp16=False)
     print(f"[2/7] Save INT8 model: {int8_ir_path}")
     int8_model_size = get_model_size(int8_ir_path, verbose=True)