IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[POT] Fixed bug with out port for nodes with multiple outputs (#10932)

Browse Source 
* Fixed bug with out port

* Add test

* Fix test

* Change test model
This commit is contained in:
Liubov Talamanova
2022-03-23 11:17:06 +03:00
committed by GitHub
parent cd361ecae1
commit 99f0093615
9 changed files with 53 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tools/pot/openvino/tools/pot/graph/node_utils.py
View File

@@ -192,8 +192,11 @@ def get_quantized_input_key(quantized_node):
Otherwise, key is tuple (fq_input name, output port number)
"""
quantized_input = get_node_input(quantized_node, 0)
quantized_key = create_node_name(quantized_input)
return quantized_key
key = quantized_input.fullname
if len(quantized_input.out_ports()) > 1:
port_number = quantized_node.in_port(0).get_source().out
key = (quantized_input.fullname, port_number)
return key
def node_with_quantized_weights(node):

29
tools/pot/openvino/tools/pot/statistics/statistic_graph_builder.py
View File

@@ -49,7 +49,7 @@ class StatisticGraphBuilder:
add_output_node, op_name = getattr(self, f'insert_{type_stat}')(model_graph,
node,
type_stat,
node.name,
node_name,
**stat.kwargs)
if add_output_node:
if node_name not in nodes_names_map[model_graph.name]:
@@ -99,13 +99,18 @@ class StatisticGraphBuilder:
axis_const = self.find_axis(node, granularity, axis)
if isinstance(axis_const, str):
return (True, node.name)
out_port = self.get_out_port(node_name)
if out_port is not None:
node_name = f'{node_name[0]}.{out_port}'
reduce_op = create_op_node_with_second_input(node.graph, insert_op, int64_array(axis_const),
dict(name=f'{type_stat}_{node_name}'))
reduce_op['fullname'] = reset_node_fullname(node.fullname, reduce_op.name)
if node.graph != model_graph:
Op.create_data_node(reduce_op.graph, reduce_op, {'shape': [1]})
node.out_port(0).connect(reduce_op.in_port(0))
return self.insert_result(model_graph, node, reduce_op, type_stat)
node.out_port(out_port if out_port else 0).connect(reduce_op.in_port(0))
return self.insert_result(model_graph, node, reduce_op, type_stat, out_port)
def insert_min(self, model_graph, node, type_stat, node_name, **kwargs):
return self.insert_reduce(model_graph, ReduceMin, node, kwargs.get('granularity'), type_stat, node_name)
@@ -127,7 +132,12 @@ class StatisticGraphBuilder:
axis_const = self.find_axis(node, kwargs.get('granularity'))
if isinstance(axis_const, str):
return (True, node.name)
abs_node = Abs(node.graph, {"name": f'abs_{node_name}'}).create_node_with_data([node.out_node(0)]).in_node(0)
out_port = self.get_out_port(node_name)
if out_port is not None:
node_name = f'{node_name[0]}.{out_port}'
abs_node = Abs(node.graph, {"name": f'abs_{node_name}'}). \
create_node_with_data([node.out_node(out_port if out_port else 0)]).in_node(0)
max_op = create_op_node_with_second_input(node.graph, ReduceMax, int64_array(axis_const),
dict(name=f'{type_stat}_{node_name}'))
@@ -135,16 +145,18 @@ class StatisticGraphBuilder:
Op.create_data_node(max_op.graph, max_op, {'shape': [1]})
max_op['fullname'] = reset_node_fullname(node.fullname, max_op.name)
abs_node.out_port(0).connect(max_op.in_port(0))
return self.insert_result(model_graph, node, max_op, type_stat)
return self.insert_result(model_graph, node, max_op, type_stat, out_port)
@staticmethod
def insert_result(model_graph, node, child_node, name):
def insert_result(model_graph, node, child_node, name, port=None):
if node.graph != model_graph:
model_graph.graph['additional_outputs'] = child_node.fullname.split('|')
res_op = AddOutputRecursive().find_and_replace_pattern(model_graph)
ie_result_name = res_op[0].name
else:
ie_result_name = f'{name}_{node.name}'
if port is not None:
ie_result_name = ie_result_name + f'.{port}'
res_op = Result(node.graph, {'name': f'Result_{ie_result_name}'}).create_node()
child_node.out_port(0).connect(res_op.in_port(0))
return (False, ie_result_name)
@@ -165,3 +177,8 @@ class StatisticGraphBuilder:
node_name_in_graph = layout_name[0] if isinstance(layout_name, tuple) else layout_name
node_name_in_graph = node_name_in_graph.replace('/pre_fq_input', '')
return node_name_in_graph
@staticmethod
def get_out_port(node_name):
out_port = node_name[1] if isinstance(node_name, tuple) else None
return out_port

3
tools/pot/tests/data/models/lstm_outs_quantization/lstm_outs_quantization.json Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:52d25d0f9bc583d15d6e24072626466cbe35db93af13aca01444c37625e61f91
size 171

3
tools/pot/tests/data/models/lstm_outs_quantization/lstm_outs_quantization.pb Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7b0bc2275947bd97cd11b20b31a69f9175c028386d9ea9ee7f6220de339772fd
size 332731

4
tools/pot/tests/data/reference_models/squeezenet1_1_example_stat_mean_pytorch.xml
View File

@@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:466cb973bbeece0f2fc187e9dd2640ccbde151d82ff52daa1628a5120713af49
size 125455
oid sha256:f96a3afe2b788ae4e460c57a4ed25da10883a553439e27ce22a726a5d4075f20
size 125489

3
tools/pot/tests/data/test_cases_refs/lstm_outs_quantization_statistics_data.json Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3a394d015d36fdbec8bdaea631a1f2c5383a7132f065a2e35bb664d5b620cb72
size 256

0
tools/pot/tests/data/test_cases_refs/statistics_data.json → tools/pot/tests/data/test_cases_refs/mobilenet-v2-pytorch_statistics_data.json
View File

2
tools/pot/tests/test_statistic_builder.py
View File

@@ -88,7 +88,7 @@ def create_(tmp_path, models, model_name, model_framework, quantization_mode,
@pytest.mark.parametrize(
'model_name, model_framework, quantization_mode, inplace_statistics, \
algorithm, preset, granularity, add_output_nodes, type_max, type_min', TEST_MODELS,
ids=['{}_{}_{}_{}_{}_{}_{}_{}_{}'.format(m[0], m[1], m[2], m[3], m[4],
ids=['{}_{}_{}_{}_{}_{}_{}_{}_{}'.format(m[0], m[1], m[2], m[3], m[4].name,
m[5], m[6], m[7], m[8], m[9]) for m in TEST_MODELS])
def test_statistics_collector_subsets(tmp_path, models, model_name, model_framework,
quantization_mode, inplace_statistics, algorithm,

19
tools/pot/tests/test_statistics_collector.py
View File

@@ -4,7 +4,6 @@
import json
import os
from copy import copy
from pathlib import Path
import pytest
import numpy as np
@@ -18,7 +17,7 @@ from openvino.tools.pot.algorithms.quantization.minmax.algorithm import MinMaxQu
from openvino.tools.pot.algorithms.quantization.bias_correction.algorithm import BiasCorrection
from .utils.config import PATHS2DATASETS_CONFIG
TEST_MODELS = [('mobilenet-v2-pytorch', 'pytorch')]
TEST_MODELS = [('mobilenet-v2-pytorch', 'pytorch'), ('lstm_outs_quantization', 'tf')]
@pytest.mark.parametrize(
'model_name, model_framework', TEST_MODELS,
@@ -57,7 +56,7 @@ def test_statistics_collector_subsets(tmp_path, models, model_name, model_framew
out = {'MinMaxQuantization': collector.get_statistics_for_algorithm('MinMaxQuantization'),
'BiasCorrection': collector.get_statistics_for_algorithm('BiasCorrection')}
refs_file = Path(__file__).parent / 'data/test_cases_refs/statistics_data.json'
refs_file = Path(__file__).parent / 'data/test_cases_refs' / f'{model_name}_statistics_data.json'
local_path = os.path.join(tmp_path, '{}_{}.json'.format(model_name, 'statistics_data'))
local_file = open(local_path, 'w')
@@ -65,17 +64,25 @@ def test_statistics_collector_subsets(tmp_path, models, model_name, model_framew
refs = json.load(file)
eps = 1e-3
local_out = copy(out)
for algo_name, algo_val in local_out.items():
local_out = {}
for algo_name, algo_val in out.items():
local_out[algo_name] = {}
for node_name, node_val in algo_val.items():
if isinstance(node_name, tuple):
name = f'{node_name[0]}.{node_name[1]}'
else:
name = node_name
local_out[algo_name][name] = {}
for stats_name, stats_val in node_val.items():
local_out[algo_name][node_name][stats_name] = [np.array(v).tolist() for v in stats_val]
local_out[algo_name][name][stats_name] = [np.array(v).tolist() for v in stats_val]
json.dump(local_out, local_file)
for algo_name, algo_val in out.items():
for node_name, node_val in algo_val.items():
for stats_name, stats_val in node_val.items():
if stats_name in ['batch_mean_param_in', 'shape']:
continue
if isinstance(node_name, tuple):
node_name = f'{node_name[0]}.{node_name[1]}'
ref_stats_vals = refs[algo_name][node_name][stats_name]
for ref_vals, vals in zip(ref_stats_vals, stats_val):
assert np.max(np.abs(np.array(ref_vals) - vals)) < eps