enable --compress_to_fp16=True by default in MO (#15488)

* enable --compress_to_fp16 by default in MO

* corrected docs, added warning if user did't specify --compress_to_fp16 explicitly

* fix failing MO unit-tests

* do not wipe out data_type if user defined it explicitly by cli argument

* updated warning message and docs

* corrected phrasing

* corrected phrasing in FP16_Compression.md

* set compress_to_fp16=False for convert tests

* leftover: set compress_to_fp16=False for convert tests

* minor correction

* print info message in main.py, som minor changes

* typos fix

* fix losing information whether arguments set by user or got from defaults

* returned back default values instead of None

* more selective correcting of test_mo_convert_pytorch.py; added test for cases when compression is enabled/disabled or left by default

* fix test_mo_convert_pytorch.py

tools/mo/openvino/tools/mo/convert_impl.py

@@ -214,12 +214,14 @@ def arguments_post_parsing(argv: argparse.Namespace):
         log.error(e)
         raise_ie_not_found()
 
-    if ('data_type' in argv and argv.data_type in ['FP16', 'half']) or \
-            ('compress_to_fp16' in argv and argv.compress_to_fp16 is True):
-        argv.data_type = 'FP32'
-        argv.compress_fp16 = True
-    else:
+    # Turn off compression only if it's disabled explicitly by --compress_to_fp16=False or --data_type=FP32.
+    # By default, in all other cases compression is enabled
+    if ('data_type' in argv and argv.data_type in ['FP32', 'float']) or \
+            ('compress_to_fp16' in argv and argv.compress_to_fp16 is False):
         argv.compress_fp16 = False
+    else:
+        argv.compress_fp16 = True
+    argv.data_type = 'FP32'  # if compression was enabled will be restored back to 'FP16' after apply_offline_transformations
 
     # This is just to check that transform key is valid and transformations are available
     check_available_transforms(parse_transform(argv.transform))

tools/mo/openvino/tools/mo/main.py

@@ -10,10 +10,9 @@ try:
     import openvino_telemetry as tm
 except ImportError:
     import openvino.tools.mo.utils.telemetry_stub as tm
-
+from openvino.tools.mo.utils.get_ov_update_message import get_compression_message
 from openvino.tools.mo.convert_impl import _convert
 from openvino.tools.mo.pipeline.common import get_ir_version
-from openvino.tools.mo.utils.cli_parser import get_model_name_from_args
 from openvino.tools.mo.utils.logger import init_logger
 from openvino.tools.mo.utils.error import Error, FrameworkError
 import traceback
@@ -38,6 +37,10 @@ def main(cli_parser: argparse.ArgumentParser, framework=None):
     try:
         ngraph_function, argv = _convert(cli_parser, framework, {})
         is_tf, _, _, _, _ = deduce_legacy_frontend_by_namespace(argv)
+        if ('compress_to_fp16' in argv and argv.compress_to_fp16) \
+                or ('data_type' in argv and argv.data_type in ['half', 'FP16']):
+            print(get_compression_message())
+
         ov_update_message = get_ov_update_message()
         ov_api20_message = get_ov_api20_message()
         if ov_update_message is not None:

tools/mo/openvino/tools/mo/utils/cli_parser.py

@@ -559,7 +559,9 @@ mo_convert_params = {
         ' used to fuse dimensions, for example "[n,c,...]->[n*c,...]".', '', '', layout_param_to_str),
     'compress_to_fp16': ParamDescription(
         'If the original model has FP32 weights or biases, they are compressed to FP16. '
-        'All intermediate data is kept in original precision.', '', '', None),
+        'All intermediate data is kept in original precision. Option can be specified alone as "--compress_to_fp16", '
+        'or explicit True/False values can be set, for example: "--compress_to_fp16=False", or "--compress_to_fp16=True"',
+        '', '', None),
     'transform': ParamDescription(
         'Apply additional transformations. {}' +
         '"--transform transformation_name1[args],transformation_name2..." ' +
@@ -1017,14 +1019,14 @@ def get_common_cli_parser(parser: argparse.ArgumentParser = None):
                                    'FP32 model weights and biases are compressed to FP16. '
                                    'All intermediate data is kept in original precision.',
                               choices=["FP16", "FP32", "half", "float"],
-                              default='float',
+                              default='FP16',
                               action=DeprecatedOptionCommon)
     common_group.add_argument('--compress_to_fp16',
                               help=mo_convert_params_common['compress_to_fp16'].description,
                               type=check_bool,
                               nargs="?",
                               const=True,
-                              default=False)
+                              default=True)
     common_group.add_argument('--transform',
                               help=mo_convert_params_common['transform'].description.format(
                                   mo_convert_params_common['transform'].possible_types_command_line),

tools/mo/openvino/tools/mo/utils/get_ov_update_message.py

@@ -32,3 +32,12 @@ def get_tf_fe_message():
               'Find more information about new TensorFlow Frontend at {}'.format(link)
 
     return message
+
+
+def get_compression_message():
+    link = "https://docs.openvino.ai/latest/openvino_docs_MO_DG_FP16_Compression.html"
+    message = '[ INFO ] Generated IR will be compressed to FP16. ' \
+              'If you get lower accuracy, please consider disabling compression explicitly ' \
+              'by adding argument --compress_to_fp16=False.\n' \
+              'Find more information about compression to FP16 at {}'.format(link)
+    return message

tools/mo/unit_tests/moc_tf_fe/check_info_messages.py

@@ -9,14 +9,15 @@ from contextlib import redirect_stdout
 from unittest.mock import patch
 
 from openvino.tools.mo.main import main
-from openvino.tools.mo.utils.get_ov_update_message import get_tf_fe_message
+from openvino.tools.mo.utils.get_ov_update_message import get_tf_fe_message, get_compression_message
 
 
 def arg_parse_helper(input_model,
                      use_legacy_frontend,
                      use_new_frontend,
                      input_model_is_text,
-                     framework):
+                     framework,
+                     compress_to_fp16=False):
     path = os.path.dirname(__file__)
     input_model = os.path.join(path, "test_models", input_model)
 
@@ -48,6 +49,7 @@ def arg_parse_helper(input_model,
         freeze_placeholder_with_value=None,
         data_type=None,
         tensorflow_custom_operations_config_update=None,
+        compress_to_fp16=compress_to_fp16,
     )
 
 
@@ -63,3 +65,18 @@ class TestInfoMessagesTFFE(unittest.TestCase):
             std_out = f.getvalue()
         tf_fe_message_found = get_tf_fe_message() in std_out
         assert tf_fe_message_found
+
+
+class TestInfoMessagesCompressFP16(unittest.TestCase):
+    @patch('argparse.ArgumentParser.parse_args',
+           return_value=arg_parse_helper(input_model="model_int32.pbtxt",
+                                         use_legacy_frontend=False, use_new_frontend=True,
+                                         compress_to_fp16=True,
+                                         framework=None, input_model_is_text=True))
+    def test_compress_to_fp16(self, mock_argparse):
+        f = io.StringIO()
+        with redirect_stdout(f):
+            main(argparse.ArgumentParser())
+            std_out = f.getvalue()
+        fp16_compression_message_found = get_compression_message() in std_out
+        assert fp16_compression_message_found