def __init__(self,
name: str,
values: Union[np.ndarray, LazyValues],
data_location: int = None):
"""
Represents a Tensor whose value is known.
Args:
name (str): The name of the tensor.
values (numpy.ndarray): The values in this tensor, in the form of a NumPy array.
data_location (int):
An enum value indicating the location where the tensor data is stored.
Generally, this will come from onnx.TensorProto.DataLocation.
"""
self.name = name
self.inputs = misc.SynchronizedList(self,
field_name="outputs",
initial=[])
self.outputs = misc.SynchronizedList(self,
field_name="inputs",
initial=[])
if not isinstance(values, np.ndarray) and not isinstance(
values, LazyValues):
G_LOGGER.critical(
"Provided `values` argument is not a NumPy array or a LazyValues instance. "
"Please provide a NumPy array or LazyValues instance to construct a Constant. "
"Note: Provided `values` parameter was: {:}".format(values))
self._values = values
self.data_location = data_location
def _get_node_id(self, node):
try:
return node.id
except AttributeError:
G_LOGGER.critical(
"Encountered a node not in the graph:\n{:}.\n\nTo fix this, please append the node to this graph's `nodes` attribute."
.format(node))
def add_to_tensor_map(tensor):
if not tensor.is_empty():
if check_duplicates and tensor.name in tensor_map and not (tensor_map[tensor.name] is tensor):
G_LOGGER.critical("Found distinct tensors that share the same name:\n[id: {:}] {:}\n[id: {:}] {:}"
.format(id(tensor_map[tensor.name]), tensor_map[tensor.name], id(tensor), tensor))
tensor_map[tensor.name] = tensor
def export_value_info_proto(tensor: Variable, do_type_check: bool) -> onnx.ValueInfoProto:
if do_type_check and tensor.dtype is None:
G_LOGGER.critical("Graph input and output tensors must include dtype information. Please set the dtype attribute for: {:}".format(tensor))
if tensor.dtype is not None:
onnx_tensor = onnx.helper.make_tensor_value_info(tensor.name, dtype_to_onnx(tensor.dtype), tensor.shape)
else:
onnx_tensor = onnx.helper.make_empty_tensor_value_info(tensor.name)
return onnx_tensor
def process_io(io):
new_io = []
for elem in io:
if isinstance(elem, Tensor):
new_io.append(elem)
elif isinstance(elem, str):
tensor = Variable(name=self._generate_name(elem))
new_io.append(tensor)
elif isinstance(elem, np.ndarray):
new_io.append(Constant(name=self._generate_name("onnx_graphsurgeon_constant"), values=elem))
else:
G_LOGGER.critical("Unrecognized type passed to Graph.layer: {:}.\n\tHint: Did you forget to unpack a list with `*`?\n\tPlease use Tensors, strings, or NumPy arrays.".format(elem))
return new_io
def __init__(self, name: str, values: np.ndarray):
"""
Represents a Tensor whose value is known.
Args:
name (str): The name of the tensor.
values (numpy.ndarray): The values in this tensor, in the form of a NumPy array.
dtype (numpy.dtype): The data type of the tensor.
shape (Sequence[Union[int, str]]): The shape of the tensor.
"""
self.name = name
self.inputs = misc.SynchronizedList(self, field_name="outputs", initial=[])
self.outputs = misc.SynchronizedList(self, field_name="inputs", initial=[])
if not isinstance(values, np.ndarray):
G_LOGGER.critical("Provided `values` argument is not a NumPy array (please provide a NumPy array to construct a Constant): {:}".format(values))
self.values = np.array(values)
def get_hierarchy_level(node):
# Return all nodes that contribute to this node.
def get_input_nodes(node):
inputs = {}
for tensor in node.inputs:
for node in tensor.inputs:
inputs[self._get_node_id(node)] = node
return inputs.values()
if self._get_node_id(node) in hierarchy_levels:
return hierarchy_levels[self._get_node_id(node)].level
# The level of a node is the level of it's highest input + 1.
try:
max_input_level = max([get_hierarchy_level(input_node) for input_node in get_input_nodes(node)] + [-1])
except RecursionError:
G_LOGGER.critical("Cycle detected in graph! Are there tensors with duplicate names in the graph?")
return max_input_level + 1
