[POT] Fixed non-zero zero point value in symmetric quantized weights (#13013)

* Fixed non-zero zero point val in symmquantized weights

* Update references

* Add tests

tools/pot/openvino/tools/pot/api/samples/classification/classification_sample.py

@@ -212,7 +212,8 @@ def optimize_model(args):
 
     # Step 7 (Optional): Compress model weights quantized precision
     #                    in order to reduce the size of final .bin file.
-    compress_model_weights(compressed_model)
+    if not args.keep_uncompressed_weights:
+        compress_model_weights(compressed_model)
 
     return compressed_model, pipeline
 

tools/pot/openvino/tools/pot/graph/passes.py

@@ -11,7 +11,7 @@ from typing import List, Set
 
 import numpy as np
 from openvino.tools.mo.back.ForceStrictPrecision import ForceStrictPrecision
-from openvino.tools.mo.back.compress_quantized_weights import CompressQuantizeWeights
+from openvino.tools.mo.back.compress_quantized_weights import CompressQuantizeWeights, ZeroPointOptimizer
 from openvino.tools.mo.ops.elementwise import Add
 from openvino.tools.mo.ops.Cast import Cast
 from openvino.tools.mo.ops.fakequantize import FakeQuantize
@@ -1013,6 +1013,8 @@ def compress_weights(model: Graph):
     add_removed_converts(model)
     CompressQuantizeWeights().find_and_replace_pattern(model)
     model.clean_up()
+    ZeroPointOptimizer().find_and_replace_pattern(model)
+    model.clean_up()
     ForceStrictPrecision().find_and_replace_pattern(model)
     model.clean_up()
 

tools/pot/tests/test_sanity.py

@@ -27,7 +27,7 @@ TEST_MODELS = [
                                                                                     'accuracy@top5': 0.907},
      {}, 'CPU'),
 
-    ('mobilenet-v2-pytorch', 'pytorch', 'DefaultQuantization', 'mixed', 300, {'accuracy@top1': 0.734,
+    ('mobilenet-v2-pytorch', 'pytorch', 'DefaultQuantization', 'mixed', 300, {'accuracy@top1': 0.737,
                                                                               'accuracy@top5': 0.91},
      {}, 'CPU'),
 
@@ -173,7 +173,7 @@ SIMPLIFIED_TEST_MODELS = [
     ('mobilenet-v2-pytorch', 'pytorch', 'DefaultQuantization', 'performance',
      {'accuracy@top1': 0.707, 'accuracy@top5': 0.91}, []),
     ('mobilenet-v2-pytorch', 'pytorch', 'DefaultQuantization', 'performance',
-     {'accuracy@top1': 0.709, 'accuracy@top5': 0.904}, ['--input_shape=[1,3,?,?]'])
+     {'accuracy@top1': 0.706, 'accuracy@top5': 0.904}, ['--input_shape=[1,3,?,?]'])
 ]