def find_and_replace_pattern(self, graph: Graph):
for dequantize_node in graph.get_op_nodes(op='DequantizeLinear'):
node_name = dequantize_node.soft_get('name', dequantize_node.id)
axis = dequantize_node.soft_get('axis', None)
scale_y_shape = dequantize_node.in_port(1).data.get_shape()
model_data_type = data_type_str_to_np(
graph.graph['cmd_params'].data_type)
cast = Cast(graph, {
'dst_type': model_data_type,
'name': node_name + '/Cast'
}).create_node()
dequantize_node.in_port(0).get_connection().set_destination(
cast.in_port(0))
mul = Mul(graph, {'can_be_fused': False}).create_node()
is_second_port_connected = dequantize_node.is_in_port_connected(2)
if is_second_port_connected:
# its is necessary not to replace subrtract for pattern in offline transformations
# See ConvertQuantizeDequantize transformation in ngraph
sub = Sub(graph, {
'name': node_name + '/Sub',
'zero_point_sub': True
}).create_node()
cast.out_port(0).connect(sub.in_port(0))
dequantize_node.in_port(2).get_connection().set_destination(
sub.in_port(1))
sub.out_port(0).connect(mul.in_port(0))
else:
cast.out_port(0).connect(mul.in_port(0))
dequantize_node.in_port(1).get_connection().set_destination(
mul.in_port(1))
dequantize_node.out_port(0).get_connection().set_source(
mul.out_port(0))
rename_nodes([(dequantize_node, node_name + '/TBD'),
(mul, node_name)])
assert scale_y_shape is not None
if axis is not None and len(
scale_y_shape) > 0 and scale_y_shape[0] > 1:
input_shape = cast.in_port(0).data.get_shape()
target_shape = np.ones(len(input_shape), np.int64)
target_shape[axis] = input_shape[axis]
mul_reshape = create_op_with_const_inputs(
graph, Reshape, {1: int64_array(target_shape)},
{'name': node_name + '/Reshape/Mul'})
mul.in_port(1).get_connection().set_destination(
mul_reshape.in_port(0))
mul_reshape.out_port(0).connect(mul.in_port(1))
if is_second_port_connected:
sub_reshape = create_op_with_const_inputs(
graph, Reshape, {1: int64_array(target_shape)},
{'name': node_name + '/Reshape/Sub'})
sub.in_port(1).get_connection().set_destination(
sub_reshape.in_port(0))
sub_reshape.out_port(0).connect(sub.in_port(1))
def find_and_replace_pattern(self, graph: Graph):
for node in graph.get_op_nodes(op='SpaceToBatch') + graph.get_op_nodes(
op='BatchToSpace'):
node.add_input_port(3, skip_if_exist=True)
# convert TF representation of the pads/crops as [N, 2] to IE representation: [N] and [N]
transposed_pads = create_op_with_const_inputs(
graph, Transpose, {1: int64_array([1, 0])})
node.in_port(2).get_connection().set_destination(
transposed_pads.in_port(0))
split_pads = create_op_with_const_inputs(graph, Split,
{1: int64_array(0)},
{'num_splits': 2})
transposed_pads.out_port(0).connect(split_pads.in_port(0))
for port_ind in range(2):
node.in_port(port_ind + 2).connect(
split_pads.out_port(port_ind))
node.in_port(port_ind + 2).get_connection().insert_node(
create_op_with_const_inputs(graph, Squeeze,
{1: int64_array([0])}))
# add zeros/ones to related inputs to align it with data input
in0_rank = Rank(graph, {
'name': node.name + '/rank_0'
}).create_node()
in1_shape = Shape(graph, {
'name': node.name + '/rank_1'
}).create_node()
diff_size = Sub(graph, {
'name': node.name + '/sub_0'
}).create_node()
diff = Sub(graph, {'name': node.name + '/sub_1'}).create_node()
const_begin = Const(graph, {
'value': int64_array([1])
}).create_node()
const_pad_val = Const(graph, {
'value': int64_array(1)
}).create_node()
block_shape = Pad(graph, {
'name': node.name + '/aligned_block_shape',
'mode': 'constant'
}).create_node()
# in case of SpaceToBatch begin = pads_begin, end = pads_end
# in case of BatchToSpace begin = crops_begin, end = crops_end
new_begin_name = '/aligned_pads_begin'
new_end_name = '/aligned_pads_end'
if node.type == 'BatchToSpace':
new_begin_name = '/aligned_crops_begin'
new_end_name = '/aligned_crops_end'
begin = Pad(graph, {
'name': node.name + new_begin_name,
'mode': 'constant'
}).create_node()
end = Pad(graph, {
'name': node.name + new_end_name,
'mode': 'constant'
}).create_node()
in0_rank_1d = create_op_node_with_second_input(
graph, Unsqueeze, int64_array([0]),
{'name': node.name + '/1d_rank_of_0'}, in0_rank)
node.in_port(0).get_source().connect(in0_rank.in_port(0))
node.in_port(1).get_source().connect(in1_shape.in_port(0))
in0_rank_1d.out_port(0).connect(diff_size.in_port(0))
in1_shape.out_port(0).connect(diff_size.in_port(1))
diff_size.out_port(0).connect(diff.in_port(0))
const_begin.out_port(0).connect(diff.in_port(1))
const_pad_val.out_port(0).connect(block_shape.in_port(3))
inputs_array = [block_shape, begin, end]
for idx, input_to_node in enumerate(inputs_array):
name_of_input_to_node = input_to_node.name
node.in_port(idx + 1).get_connection().set_destination(
input_to_node.in_port(0))
const_begin.out_port(0).connect(input_to_node.in_port(1))
diff.out_port(0).connect(input_to_node.in_port(2))
input_to_node.out_port(0).connect(node.in_port(idx + 1))
convert = Cast(graph, {
'name': name_of_input_to_node + '/i64',
'dst_type': np.int64
}).create_node()
input_to_node.in_port(0).get_connection().insert_node(convert)
def dequantize_data(fake_quantize: Node, dst_type: type,
quantized_type: type) -> Node:
graph = fake_quantize.graph
quantized_data = fake_quantize.in_port(0).get_source().node
name = fake_quantize.soft_get('name', fake_quantize.id)
assert quantized_data.soft_get('type') == 'Convert' and quantized_data.dst_type == quantized_type, \
'Weights aren`t compressed as expected for node {}'.format(fake_quantize.soft_get('name', fake_quantize.id))
dequantizing_cast = Cast(
graph,
dict(name=quantized_data.name +
"/to_{}".format(np_data_type_to_destination_type(dst_type)),
dst_type=dst_type,
stop_value_propagation=True)).create_node()
fake_quantize.in_port(0).get_connection().set_destination(
dequantizing_cast.in_port(0))
# limits of dequantize
in_low = fake_quantize.in_port(1).get_source()
in_high = fake_quantize.in_port(2).get_source()
out_low = fake_quantize.in_port(3).get_source()
out_high = fake_quantize.in_port(4).get_source()
# scale calculation
output_range = Sub(graph, {
'name': name + '/output_range'
}).create_node()
output_range.in_port(0).connect(out_high)
output_range.in_port(1).connect(out_low)
input_range = Sub(graph, {'name': name + '/input_range'}).create_node()
input_range.in_port(0).connect(in_high)
input_range.in_port(1).connect(in_low)
scale = Div(graph, {'name': name + '/scale'}).create_node()
scale.in_port(0).connect(output_range.out_port(0))
scale.in_port(1).connect(input_range.out_port(0))
# shift calculation
descaled_output_low = Div(graph, {
'name': name + '/descaled_output_low'
}).create_node()
descaled_output_low.in_port(0).connect(out_low)
descaled_output_low.in_port(1).connect(scale.out_port(0))
shift = Sub(graph, {'name': name + '/shift'}).create_node()
shift.in_port(0).connect(in_low)
shift.in_port(1).connect(descaled_output_low.out_port(0))
zero = Const(graph, {
'name': name + '/zero',
'value': mo_array(0, dtype=dst_type)
}).create_node()
scale_eq_zero = Equal(graph, {
'name': name + '/scale_eq_zero'
}).create_node()
scale_eq_zero.in_port(0).connect(scale.out_port(0))
scale_eq_zero.in_port(1).connect(zero.out_port(0))
zero_point = Select(graph, {
'name': name + '/zero_point'
}).create_node()
zero_point.in_port(0).connect(scale_eq_zero.out_port(0))
zero_point.in_port(1).connect(zero.out_port(0))
zero_point.in_port(2).connect(shift.out_port(0))
# DeQuantize(x) == Mul(Sub(x, zero_point), scale)
sub_zp = Sub(graph, {'name': name + '/minus_zp'}).create_node()
sub_zp.in_port(0).connect(dequantizing_cast.out_port(0))
sub_zp.in_port(1).connect(zero_point.out_port(0))
mul_scale = Mul(graph, {
'name': name + '/mulpiply_by_scale'
}).create_node()
mul_scale.in_port(0).connect(sub_zp.out_port(0))
mul_scale.in_port(1).connect(scale.out_port(0))
fake_quantize.out_port(0).get_connection().set_source(
mul_scale.out_port(0))
graph.remove_nodes_from([fake_quantize.id, fake_quantize.out_node(0)])
def mxrepeat_decomposition(node: Node):
graph = node.graph
name = node.soft_get('name', node.id)
rename_node(node, name + '/to_be_removed')
# Unqueeze
input_rank = Rank(graph, {'name': name + '/Rank'}).create_node()
node.in_port(0).get_source().connect(input_rank.in_port(0))
axis = get_canonical_axis_index_node(input_rank, node.axis)
unsqueeze_axis = create_op_node_with_second_input(
graph,
Add,
int64_array([1]), {'name': name + '/Unsqueeze/Axis'},
input_node=axis)
unsqueeze = Unsqueeze(graph, {
'name': name + '/Unsqueeze'
}).create_node()
unsqueeze.in_port(1).connect(unsqueeze_axis.out_port(0))
# Tile (1, 1, ..., repeats, ..., 1)
# we generate tile array according to the following table:
# parts: | first | repeats | second |
# i: | 0, 1, ..., axis,| axis + 1,| ..., rank+1 |
# tile_array: | 1, 1, ..., 1 ,| repeats ,| ..., 1 |
one = Const(graph, {
'name': name + '/Broadcast/One',
'value': int64_array([1])
}).create_node()
first_ones = Broadcast(graph, {
'name': name + '/Broadcast/Ones_first_part'
}).create_node()
first_ones.in_port(0).connect(one.out_port(0))
first_ones.in_port(1).connect(unsqueeze_axis.out_port(0))
repeats = Const(graph, {
'name': name + '/repeats',
'value': int64_array([node.repeats])
}).create_node()
second_ones = Broadcast(graph, {
'name': name + '/Broadcast/Ones_second_part'
}).create_node()
second_part_broadcast_shape = Sub(
graph, {
'name': name + '/Broadcast/Shape/second_part'
}).create_node()
second_part_broadcast_shape.in_port(0).connect(input_rank.out_port(0))
second_part_broadcast_shape.in_port(1).connect(
unsqueeze_axis.out_port(0))
second_ones.in_port(0).connect(one.out_port(0))
second_ones.in_port(1).connect(second_part_broadcast_shape.out_port(0))
tile_repeats = new_shape_node_from_shape_nodes(
[first_ones, repeats, second_ones])
tile = Tile(graph, {'name': name + '/Tile'}).create_node()
tile.in_port(1).connect(tile_repeats.out_port(0))
# Reshape (input_shape[:axis], input_shape[axis] * repeats, input_shape[axis+1:])
# we generate reshape dim array according to the following table:
# parts: | first | rep | second |
# i: | 0, 1, ... ,| axis, | ..., rank |
# dim_array: | inp_sh[i] ,| input_shape[axis] * repeats ,| inp_sh[i] |
input_shape = Shape(graph, {'name': name + '/Shape'}).create_node()
node.in_port(0).get_source().connect(input_shape.in_port(0))
first_input_shape_part = get_shape_values_by_range_idxs(
input_shape,
input_rank,
begin=0,
end=node.axis,
include_begin=True,
include_end=False)
original_axis_dim = create_op_with_const_inputs(
graph,
Gather, {2: int64_array(0)}, {'name': name + '/OriginalDim'},
input_node=input_shape)
original_axis_dim.in_port(1).connect(axis.out_port(0))
repeated_dimention = Mul(graph, {
'name': name + '/RepeatedDim'
}).create_node()
repeated_dimention.in_port(0).connect(original_axis_dim.out_port(0))
repeated_dimention.in_port(1).connect(repeats.out_port(0))
second_input_shape_part = get_shape_values_by_range_idxs(
input_shape,
input_rank,
begin=node.axis,
end=-1,
include_begin=False,
include_end=True)
output_shape = new_shape_node_from_shape_nodes([
first_input_shape_part, repeated_dimention, second_input_shape_part
])
reshape = Reshape(graph, {'name': name}).create_node()
rename_node(reshape, name)
reshape.in_port(1).connect(output_shape.out_port(0))
# Final connections
node.in_port(0).get_connection().set_destination(unsqueeze.in_port(0))
tile.in_port(0).connect(unsqueeze.out_port(0))
reshape.in_port(0).connect(tile.out_port(0))
node.out_port(0).get_connection().set_source(reshape.out_port(0))