def prepare_emit_ir(graph: nx.MultiDiGraph, data_type: str, output_dir: str, output_model_name: str,
mean_data: [list, None] = None, input_names: list = [], meta_info: dict = dict()):
for sub_graph in [graph] + collect_sub_graphs(graph):
create_const_nodes(sub_graph, start_data_nodes_are_not_allowed=(sub_graph == graph))
op_order, data_order = determined_sort(get_sorted_outputs(sub_graph))
mapping = {v: u for u, v in enumerate(op_order)}
mapping.update({v: u for u, v in enumerate(data_order, start=len(sub_graph))})
relabel_nodes_inplace_safe(sub_graph, mapping)
port_renumber(sub_graph)
convert_data_type.convert(sub_graph, data_type)
tensor_names.propagate_op_name_to_tensor(graph)
weights = np.array([])
bin_file = os.path.join(output_dir, '{}.bin'.format(output_model_name))
if(data_type == "FP16"):
weights = serialize_constants(weights,graph,data_type=np.float16)
elif(data_type == "FP32"):
weights = serialize_constants(weights,graph,data_type=np.float32)
mean_offset = None
mean_size = None
if mean_data:
mean_offset, mean_size = serialize_mean_image(bin_file, mean_data=mean_data)
xml_string = generate_ie_ir(graph=graph,
file_name=os.path.join(output_dir, '{}.xml'.format(output_model_name)),
input_names=input_names,
mean_offset=mean_offset,
mean_size=mean_size,
meta_info=meta_info)
# tensor_names.output_tensor_names_map(graph, os.path.join(output_dir, '{}.mapping'.format(output_model_name)))
return weights, xml_string
def test_convert_blob_to_int32_with_force_precision_error(self):
graph = build_graph(
nodes_attributes, [('data_node', 'op_node', {
'bin': 1
})], {
'data_node': {
'force_precision': 'int32',
'value': np.array([4.0, 3.0, 2.0, 1.1], dtype=np.float64)
}
})
with self.assertRaisesRegex(Error, '.*results in rounding.*'):
convert(graph, "FP32")
def test_convert_blob_to_fp16_from_fp64_overflow(self):
graph = build_graph(
nodes_attributes, [('data_node', 'op_node', {
'bin': 1
})], {
'data_node': {
'value': np.array([4.0, 3.0, 2.0, 1e10], dtype=np.float64)
}
})
convert(graph, "FP16")
result_value = graph.node['data_node']['value']
self.assertTrue(result_value.dtype == np.float16)
self.assertListEqual(list(result_value), [4, 3, 2, np.inf])
def test_convert_blob_to_int32_with_force_precision(self):
graph = build_graph(
nodes_attributes, [('data_node', 'op_node', {
'bin': 1
})], {
'data_node': {
'force_precision': 'int32',
'value': np.array([4.0, 3.0, 2.0, 1.0], dtype=np.float64)
}
})
convert(graph, "FP32")
result_value = graph.node['data_node']['value']
self.assertTrue(result_value.dtype == np.int32)
self.assertListEqual(list(result_value), [4, 3, 2, 1])
def prepare_emit_ir(graph: Graph,
data_type: str,
output_dir: str,
output_model_name: str,
mean_data: [list, None] = None,
input_names: list = [],
meta_info: dict = dict()):
graph.strict_mode = False
for sub_graph in [graph] + collect_sub_graphs(graph):
op_order, data_order = determined_sort(get_sorted_outputs(sub_graph))
mapping = {v: u for u, v in enumerate(op_order)}
mapping.update(
{v: u
for u, v in enumerate(data_order, start=len(sub_graph))})
relabel_nodes_inplace_safe(sub_graph, mapping)
port_renumber(sub_graph)
convert_data_type.convert(sub_graph, data_type)
tensor_names.propagate_op_name_to_tensor(graph)
bin_file = os.path.join(output_dir, '{}.bin'.format(output_model_name))
serialize_constants(graph, bin_file)
mean_offset = None
mean_size = None
if mean_data:
mean_offset, mean_size = serialize_mean_image(bin_file,
mean_data=mean_data)
generate_ie_ir(graph=graph,
file_name=os.path.join(output_dir,
'{}.xml'.format(output_model_name)),
input_names=input_names,
mean_offset=mean_offset,
mean_size=mean_size,
meta_info=meta_info)
tensor_names.output_tensor_names_map(
graph, os.path.join(output_dir,
'{}.mapping'.format(output_model_name)))
