Files
openvino/model-optimizer/extensions/back/remove_last_softmax_pattern.py

130 lines
5.3 KiB
Python

"""
Copyright (C) 2018-2020 Intel Corporation
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
"""
import logging as log
from mo.back.replacement import BackReplacementPattern
from mo.front.common.partial_infer.utils import int64_array
from mo.graph.graph import Graph
from mo.middle.passes.eliminate import remove_op_node_with_data_node
class RemoveLastSoftMaxPattern(BackReplacementPattern):
enabled = True
graph_condition = [lambda graph: graph.graph['fw'] == 'kaldi' and graph.graph['cmd_params'].remove_output_softmax]
@staticmethod
def pattern():
return dict(
nodes=[
('softmax_node', dict(op='SoftMax')),
('softmax_data', dict(kind='data')),
('op_output', dict(op='Result'))
],
edges=[
('softmax_node', 'softmax_data'),
('softmax_data', 'op_output')
]
)
@staticmethod
def replace_pattern(graph: Graph, match: dict):
"""
Removes output SoftMax layer
:param graph: graph to operate on
:param match: dictionary with matched nodes
"""
if len(match['softmax_data'].out_nodes()) == 1:
remove_op_node_with_data_node(graph, match['softmax_node'])
else:
log.error("SoftMax is not last layer, so can't be removed", extra={'is_warning': True})
class RemoveLastLogSoftMaxPattern(BackReplacementPattern):
enabled = True
graph_condition = [lambda graph: graph.graph['fw'] == 'kaldi' and graph.graph['cmd_params'].remove_output_softmax]
force_clean_up = True
@staticmethod
def pattern():
return dict(
nodes=[
('input_data', {'kind': 'data'}),
('sub_node', {'kind': 'op', 'op': 'Sub'}),
('reduce_max_node', {'kind': 'op', 'op': 'ReduceMax'}),
('reduce_max_node_data', {'kind': 'data'}),
('sub_node_data', {'kind': 'data'}),
('exp', {'kind': 'op', 'op': 'Exp'}),
('exp_data', {'kind': 'data'}),
('reduce_sum_node', {'kind': 'op', 'op': 'ReduceSum'}),
('reduce_sum_node_data', {'kind': 'data'}),
('reduce_sum_axis', {'kind': 'op', 'op': 'Const'}),
('reduce_sum_axis_data', {'kind': 'data'}),
('log', {'kind': 'op', 'op': 'Log'}),
('log_data', {'kind': 'data'}),
('last_sub', {'kind': 'op', 'op': 'Sub'}),
('last_sub_data', {'kind': 'data'}),
('op_output', {'kind': 'op', 'op': 'Result'}),
],
edges=[
('input_data', 'sub_node', {'in': 0}),
('input_data', 'reduce_max_node', {'in': 0}),
('reduce_max_node', 'reduce_max_node_data'),
('reduce_max_node_data', 'sub_node', {'in': 1}),
('sub_node', 'sub_node_data'),
('sub_node_data', 'exp', {'out': 0, 'in': 0}),
('exp', 'exp_data'),
('exp_data', 'reduce_sum_node', {'in': 0}),
('reduce_sum_node', 'reduce_sum_node_data'),
('reduce_sum_axis', 'reduce_sum_axis_data'),
('reduce_sum_axis_data', 'reduce_sum_node', {'in': 1}),
('reduce_sum_node_data', 'log'),
('log', 'log_data'),
('log_data', 'last_sub', {'in': 1}),
('last_sub', 'last_sub_data'),
('sub_node_data', 'last_sub', {'out': 0, 'in': 0}),
('last_sub_data', 'op_output'),
]
)
expected_number_of_outputs = {
'reduce_max_node': 1, 'reduce_sum_node': 1, 'exp': 1, 'log': 1, 'sub_node': 2, 'last_sub': 1
}
@staticmethod
def replace_pattern(graph: Graph, match: dict):
"""
Removes output LogSoftMax layer
:param graph: graph to operate on
:param match: dictionary with matched nodes
"""
reduce_max_node = match['reduce_max_node']
second_input_of_reduce_max = reduce_max_node.in_port(1).get_connection().get_source().node
if not second_input_of_reduce_max.has_valid('value') or len(second_input_of_reduce_max.value) != 1:
return
reduce_sum_node = match['reduce_sum_node']
second_input_of_reduce_sum = reduce_sum_node.in_port(1).get_connection().get_source().node
if not second_input_of_reduce_sum.has_valid('value') or len(second_input_of_reduce_sum.value) != 1:
return
if second_input_of_reduce_max.value[0] != second_input_of_reduce_sum.value[0]:
return
for name, number in RemoveLastLogSoftMaxPattern.expected_number_of_outputs.items():
if len(match[name].out_port(0).get_destinations()) != number:
return
match['op_output'].in_port(0).get_connection().set_source(match['sub_node'].in_port(0).get_source())