[ MO ] Keep data type of compressed value (#1143)

JIRA: 34085

diff --git a/model-optimizer/extensions/back/compress_quantized_weights.py b/model-optimizer/extensions/back/compress_quantized_weights.py
index 2bd2e40..df2500c 100644 (file)
--- a/model-optimizer/extensions/back/compress_quantized_weights.py
+++ b/model-optimizer/extensions/back/compress_quantized_weights.py
@@ -98,14 +98,10 @@ class CompressQuantizeWeights(BackReplacementPattern):
                 ('weights_const', dict(type='Const')),
                 ('weights_d', dict(kind='data')),
                 ('quantize', dict(type='FakeQuantize', levels=lambda x: x is not None and 2 < x <= 256)),
-                ('quantize_d', dict(kind='data')),
-                ('convolution', dict())
             ],
             edges=[
                 ('weights_const', 'weights_d'),
                 ('weights_d', 'quantize', {'in': 0}),
-                ('quantize', 'quantize_d'),
-                ('quantize_d', 'convolution', {'in': 1})
             ]
         )
 
@@ -118,7 +114,9 @@ class CompressQuantizeWeights(BackReplacementPattern):
         initial_fake_quantize.in_port(1).get_connection().set_destination(new_fake_quantize.in_port(1))
         initial_fake_quantize.in_port(2).get_connection().set_destination(new_fake_quantize.in_port(2))
 
-        dst_type = data_type_str_to_np(graph.graph['cmd_params'].data_type)
+        dst_type = match['weights_const'].value.dtype
+        if np.issubdtype(dst_type, np.floating):
+            dst_type = data_type_str_to_np(graph.graph['cmd_params'].data_type)
 
         i_min = np.array([0.], dtype=dst_type)
         i_max = np.array([initial_fake_quantize.levels - 1.], dtype=dst_type)

diff --git a/model-optimizer/extensions/back/compress_quantized_weights_test.py b/model-optimizer/extensions/back/compress_quantized_weights_test.py
index eda2e77..dedd4ea 100644 (file)
--- a/model-optimizer/extensions/back/compress_quantized_weights_test.py
+++ b/model-optimizer/extensions/back/compress_quantized_weights_test.py
@@ -13,11 +13,11 @@
  See the License for the specific language governing permissions and
  limitations under the License.
 """
-
 import unittest
 from argparse import Namespace
 
 import numpy as np
+from generator import generator, generate
 
 from extensions.back.compress_quantized_weights import CompressQuantizeWeights
 from extensions.ops.fakequantize import FakeQuantize
@@ -25,7 +25,8 @@ from mo.front.common.partial_infer.eltwise import eltwise_infer
 from mo.graph.graph import Node
 from mo.ops.const import Const
 from mo.utils.ir_engine.compare_graphs import compare_graphs
-from mo.utils.unittest.graph import build_graph
+from mo.utils.unittest.graph import build_graph, regular_op_with_shaped_data, regular_op_with_empty_data, \
+    valued_const_with_data, result, connect
 
 nodes_attributes = {
     # placeholder
@@ -33,7 +34,7 @@ nodes_attributes = {
     'placeholder_data': {'kind': 'data'},
 
     # weights
-    'weights_const': {'type': 'Const', 'kind': 'op'},
+    'weights_const': {'type': 'Const', 'kind': 'op', 'value': np.array([], dtype=np.float32)},
     'weights_data': {'kind': 'data'},
 
     # quantize
@@ -446,7 +447,6 @@ class WeightQuantizeTest(unittest.TestCase):
         self.assertTrue(flag, resp)
 
     def test_accuracy_tensor1(self):
-
         """
         [1.0, 2.0, 3.0, 4.0]
         """
@@ -814,3 +814,63 @@ class WeightQuantizeTest(unittest.TestCase):
         w_array_ref = Node(graph_ref, 'ac_weights').out_port(0).get_destination().data.get_value()
 
         self.assertTrue(np.all(w_array == w_array_ref))
+
+
+@generator
+class CompressionDataTypeTest(unittest.TestCase):
+    @generate(*[
+        ('FP32', np.int64),
+        ('FP16', np.int64),
+        ('FP32', np.int32),
+        ('FP16', np.int32),
+        ('FP32', np.float64, np.float32),
+        ('FP16', np.float64, np.float16),
+        ('FP32', np.float32, np.float32),
+        ('FP16', np.float32, np.float16),
+        ('FP32', np.float16, np.float32),
+        ('FP16', np.float16, np.float16),
+        ])
+    def test_data_type(self, model_dtype, original, transformed=None):
+        if transformed is None:
+            transformed = original
+
+        nodes = {
+            **valued_const_with_data('weights', np.ones([1, 2, 3, 4], dtype=original)),
+
+            **valued_const_with_data('int_weights', np.ones([1, 2, 3, 4], dtype=np.uint8)),
+            **regular_op_with_shaped_data('cast', [1, 2, 3, 4], {'type': 'Convert', 'dst_type': transformed}),
+
+            **valued_const_with_data('il', np.array([0])),
+            **valued_const_with_data('ih', np.array([254])),
+            **valued_const_with_data('ol', np.array([0])),
+            **valued_const_with_data('oh', np.array([254])),
+
+            **regular_op_with_shaped_data('FQ', [1, 2, 3, 4], {'type': 'FakeQuantize', 'infer': FakeQuantize.infer,
+                                                               'stop_value_propagation': True, 'levels': 255}),
+            **result()
+        }
+
+        graph = build_graph(nodes, [
+            *connect('weights:0', '0:FQ'),
+            *connect('il:0', '1:FQ'),
+            *connect('ih:0', '2:FQ'),
+            *connect('ol:0', '3:FQ'),
+            *connect('oh:0', '4:FQ'),
+            *connect('FQ:0', 'output'),
+        ], nodes_with_edges_only=True)
+        graph.graph['cmd_params'] = Namespace(data_type=model_dtype, keep_shape_ops=True)
+
+        CompressQuantizeWeights().find_and_replace_pattern(graph)
+        graph.clean_up()
+
+        graph_ref = build_graph(nodes, [
+            *connect('int_weights:0', '0:cast'),
+            *connect('cast:0', '0:FQ'),
+            *connect('il:0', '1:FQ'),
+            *connect('ih:0', '2:FQ'),
+            *connect('ol:0', '3:FQ'),
+            *connect('oh:0', '4:FQ'),
+            *connect('FQ:0', 'output'),
+        ], nodes_with_edges_only=True)
+        (flag, resp) = compare_graphs(graph, graph_ref, 'output', check_op_attrs=True)
+        self.assertTrue(flag, resp)