replied to comments

openvinotoolkit · Oct 25, 2023 · 8fce7a7 · 8fce7a7
1 parent 3056252
commit 8fce7a7
Show file tree

Hide file tree

Showing 7 changed files with 11 additions and 7 deletions.
diff --git a/examples/post_training_quantization/onnx/mobilenet_v2/main.py b/examples/post_training_quantization/onnx/mobilenet_v2/main.py
@@ -16,7 +16,7 @@
 
 import numpy as np
 import onnx
-import openvino.runtime as ov
+import openvino as ov
 import torch
 from fastdownload import FastDownload
 from fastdownload import download_url

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
@@ -19,7 +19,7 @@
 from typing import Any, Dict, Iterable, List, Optional, Tuple
 
 import numpy as np
-import openvino.runtime as ov
+import openvino as ov
 import torch
 from anomalib.data.mvtec import MVTec
 from anomalib.data.utils import download
@@ -169,6 +169,10 @@ def transform_fn(data_item):
 print(f"[1/7] Save FP32 model: {fp32_ir_path}")
 fp32_size = get_model_size(fp32_ir_path, verbose=True)
 
+# To avoid an accuracy drop when saving a model due to compression of unquantized
+# weights to FP16, compress_to_fp16=False should be used. This is necessary because
+# nncf.quantize_with_accuracy_control(...) keeps the most impactful operations within
+# the model in the original precision to achieve the specified model accuracy.
 int8_ir_path = f"{ROOT}/stfpm_int8.xml"
 ov.save_model(ov_quantized_model, int8_ir_path, compress_to_fp16=False)
 print(f"[2/7] Save INT8 model: {int8_ir_path}")

diff --git a/examples/post_training_quantization/openvino/mobilenet_v2/main.py b/examples/post_training_quantization/openvino/mobilenet_v2/main.py
@@ -16,7 +16,7 @@
 from typing import List, Optional
 
 import numpy as np
-import openvino.runtime as ov
+import openvino as ov
 import torch
 from fastdownload import FastDownload
 from sklearn.metrics import accuracy_score

diff --git a/examples/post_training_quantization/openvino/yolov8/main.py b/examples/post_training_quantization/openvino/yolov8/main.py
@@ -14,7 +14,7 @@
 from typing import Any, Dict, Tuple
 
 import numpy as np
-import openvino.runtime as ov
+import openvino as ov
 import torch
 from tqdm import tqdm
 from ultralytics.cfg import get_cfg

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
@@ -15,7 +15,7 @@
 from typing import Any, Dict, Tuple
 
 import numpy as np
-import openvino.runtime as ov
+import openvino as ov
 import torch
 from tqdm import tqdm
 from ultralytics.cfg import get_cfg

diff --git a/examples/post_training_quantization/torch/mobilenet_v2/main.py b/examples/post_training_quantization/torch/mobilenet_v2/main.py
@@ -16,7 +16,7 @@
 from typing import List, Optional
 
 import numpy as np
-import openvino.runtime as ov
+import openvino as ov
 import torch
 from fastdownload import FastDownload
 from openvino.tools import mo

diff --git a/examples/post_training_quantization/torch/ssd300_vgg16/main.py b/examples/post_training_quantization/torch/ssd300_vgg16/main.py
@@ -18,7 +18,7 @@
 import nncf
 from nncf.torch import disable_tracing
 
-import openvino.runtime as ov
+import openvino as ov
 import torch
 import torchvision
 from fastdownload import FastDownload