[POT] Update for the Results quantization & tests (openvinotoolkit#8324)

* Update rule for multi-results quantization

(cherry picked from commit d94fe7d)

* Update tests with lstm cases

(cherry picked from commit 16fe385)

* Update lstm reference

* Fix pylint issue

tools/pot/openvino/tools/pot/algorithms/quantization/fake_quantize.py

@@ -441,7 +441,12 @@ def _custom_broadcast(arrays_list):
 def create_renamed_layers_mapping(model, stats_layout):
     changed_names_map = {}
     for layer_name in stats_layout:
-        node = get_node_by_name(model, layer_name)
+        node_name = layer_name
+        port_id = None
+        if isinstance(layer_name, tuple):
+            node_name, port_id = layer_name
+        node = get_node_by_name(model, node_name)
         if node is not None and 'orig_node_name' in node:
-            changed_names_map[node.name] = node['orig_node_name']
+            name_change_to = node['orig_node_name'] if port_id is None else (node['orig_node_name'], port_id)
+            changed_names_map[layer_name] = name_change_to
     return changed_names_map

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

@@ -649,12 +649,20 @@ def rename_fqs_in_the_end(self, graph: Graph):
         def change_names(_, match):
             fq_node = match['fq']
             input_node = get_node_input(fq_node, 0)
+            new_fq_name = copy(input_node.name)
+            if 'orig_node_name' in input_node:
+                new_fq_name = copy(input_node['orig_node_name'])
+
+            input_node_outputs = get_all_node_outputs(input_node)
+            if all([op.type == 'FakeQuantize' for op in input_node_outputs]):
+                new_fq_name += '.{}'.format(fq_node.in_port(0).get_source().idx)
 
             fq_node['orig_fq_name'] = copy(fq_node.name)
-            fq_node.name = copy(input_node.name)
+            fq_node.name = copy(new_fq_name)
 
-            input_node['orig_node_name'] = copy(input_node.name)
-            input_node.name = '{original_name}/pre_fq_input'.format(original_name=input_node.name)
+            if 'orig_node_name' not in input_node:
+                input_node['orig_node_name'] = copy(input_node.name)
+                input_node.name = '{original_name}/pre_fq_input'.format(original_name=input_node.name)
 
         pattern = get_fq_result_pattern()
         apply_pattern(

tools/pot/tests/test_graph.py

@@ -18,13 +18,14 @@
 GNA_CONFIG_PATH = HARDWARE_CONFIG_PATH / 'gna.json'
 
 TEST_MODELS = [
-    ('mobilenetv2_example', 'pytorch'),
-    ('resnet_example', 'pytorch'),
-    ('googlenet_example', 'pytorch'),
-    ('mobilenetv2_ssd_example', 'pytorch'),
-    ('densenet121_example', 'pytorch'),
-    ('multiple_out_ports_net', 'tf'),  # multiple output ports in node case check,
-    # ('rm_nnet4a', 'kaldi')
+    ('mobilenetv2_example', 'pytorch', 'ANY'),
+    ('resnet_example', 'pytorch', 'ANY'),
+    ('googlenet_example', 'pytorch', 'ANY'),
+    ('mobilenetv2_ssd_example', 'pytorch', 'ANY'),
+    ('densenet121_example', 'pytorch', 'ANY'),
+    ('multiple_out_ports_net', 'tf', 'ANY'),
+    ('lstm_example', 'pytorch', 'GNA'),
+    ('multiple_outputs_net_example', 'dldt', 'GNA')
 ]
 
 CASCADE_MAP = Dict({
@@ -37,15 +38,16 @@
 
 
 @pytest.mark.parametrize(
-    'model_name, model_framework', TEST_MODELS,
+    'model_name, model_framework, target_device', TEST_MODELS,
     ids=['{}_{}'.format(m[0], m[1]) for m in TEST_MODELS])
-def test_build_quantization_graph(tmp_path, models, model_name, model_framework):
+def test_build_quantization_graph(tmp_path, models, model_name, model_framework, target_device):
     model = models.get(model_name, model_framework, tmp_path)
-    model = load_model(model.model_params)
+    model = load_model(model.model_params, target_device=target_device)
 
-    hardware_config = HardwareConfig.from_json(CPU_CONFIG_PATH.as_posix())
-    if model_framework == 'kaldi':
+    if target_device == 'GNA':
         hardware_config = HardwareConfig.from_json(GNA_CONFIG_PATH.as_posix())
+    else:
+        hardware_config = HardwareConfig.from_json(CPU_CONFIG_PATH.as_posix())
 
     quantization_model = GraphTransformer(hardware_config).insert_fake_quantize(model)
 
@@ -246,27 +248,21 @@ def test_multibranch_propagation_without_fq_moving():
 
 
 MODELS_WITH_LSTM = [
-    # ('rm_lstm4f', 'kaldi', {
-    #     'prev_memory_output69':
-    #         ['next_lstm_output108', 'lstmprojectedstreams/Shape', 'input_fullyconnected/WithoutBiases'],
-    #     'prev_memory_state82':
-    #         ['state_filtered_tahn100', 'clamp_scaleshift101/Mul_', 'next_lstm_state98'],
-    #     'prev_memory_output':
-    #         ['next_lstm_output', 'affinetransform/WithoutBiases'],
-    #     'prev_memory_state':
-    #         ['state_filtered_tahn', 'clamp_scaleshift/Mul_', 'next_lstm_state']
-    # })
+    ('lstm_example', 'pytorch', {
+        'LSTM_15/TensorIterator/22/variable_1':
+            ['Assign_298'],
+        'LSTM_15/TensorIterator/24/variable_2':
+            ['Assign_305'],
+        'LSTM_19/TensorIterator/22/variable_1':
+            ['Assign_327'],
+        'LSTM_19/TensorIterator/24/variable_2':
+            ['Assign_334']
+    })
 ]
 
 
-@pytest.fixture(scope='module', params=MODELS_WITH_LSTM,
-                ids=['{}_{}'.format(m[0], m[1]) for m in MODELS_WITH_LSTM])
-def _params(request):
-    return request.param
-
-
-def test_lstm_ends(_params, tmp_path, models):
-    model_name, model_framework, lstm_ends_ref = _params
+def test_lstm_ends(tmp_path, models):
+    model_name, model_framework, lstm_ends_ref = MODELS_WITH_LSTM[0]
     model = models.get(model_name, model_framework, tmp_path)
     model = load_model(model.model_params)
     read_values = get_nodes_by_type(model, ['ReadValue'])