def make_const_node(data: Tensor, name: str = None) -> NodeDef:
"""
Create a TF graph node containing a constant value.
The resulting node is equivalent to using `tf.constant` on the
default graph.
Args:
data: Numpy-array containing the data, shape, and datatype
name: Optional name of the node
Returns:
Graph node for adding to a TF Graph instance
"""
dtype = as_dtype(data.dtype).as_datatype_enum
tensor_content = data.tobytes()
tensor_dim = [TensorShapeProto.Dim(size=size) for size in data.shape]
tensor_shape = TensorShapeProto(dim=tensor_dim)
tensor_proto = TensorProto(tensor_content=tensor_content,
tensor_shape=tensor_shape,
dtype=dtype)
node_def = NodeDef(op='Const',
name=name or 'Const',
attr={
'value': AttrValue(tensor=tensor_proto),
'dtype': AttrValue(type=dtype)
})
return node_def
def update_colocation_group(self, get_colocation_op):
"""
Update operations colocated with master variables to be colocated with proxy variables.
Args:
get_colocation_op (Callable): fn that gets the current colocation ops
"""
for op in self._graph_item.graph.get_operations():
# Do not update shared node (including nodes in the optimizer)
# Do not update operations within the variable scope of master var
# Do not update the VarhandleOp itself
if not op.name.startswith(AUTODIST_REPLICA_PREFIX) or \
op.name.startswith(self._optimizer_name_scope) or \
op.name.startswith(self._this_op.name + '/') or \
(op.name.startswith(self._this_op.name) and op.type == 'VarHandleOp'):
continue
new_colocation_group = []
for colocation_group in op.colocation_groups():
current_binding_op = get_colocation_op(colocation_group)
if current_binding_op == self._this_op:
op_name_to_bind_to = (
COLOCATION_PREFIX +
as_bytes(self._proxy_vars[0].op.name))
new_colocation_group.append(op_name_to_bind_to)
else:
new_colocation_group.append(colocation_group)
op._set_attr(
"_class",
pb2_AttrValue(list=pb2_AttrValue.ListValue(
s=new_colocation_group)))
def make_op_node(op_name: Text, inputs: Inputs, name: Text = None) -> NodeDef:
"""
Create a TF graph node given the operation, input, and a name.
The resulting node definition won't include any operation-specific
attributes. It returns a valid node for most operations, though.
Args:
op_name: Native TF operation name (e.g. "MatMul")
inputs: Input node, node name, or list of inputs nodes or node names
name: Node name in the graph, must be unique and defaults to the
operation name
Returns:
TF graph node definition for the given operation, inputs, and name
"""
input_list = inputs
# convert scalar input into list
if not isinstance(inputs, list):
input_list = [input_list]
# convert list items to strings
for i, item in enumerate(input_list):
if hasattr(item, 'name'):
input_list[i] = item.name
# generate node defintion
dtype = dtypes.float32.as_datatype_enum
node_def = NodeDef(op=op_name,
name=name or op_name,
attr={'T': AttrValue(type=dtype)})
node_def.input.extend(input_list)
return node_def
def _prune_colocation_groups(graph_item):
for op in graph_item.graph.get_operations():
# Now prune the graph to have the right colocation constraints
colocation_groups = [(c, graph_item.get_colocation_op(c))
for c in op.colocation_groups()]
# We don't want any colocation groups that are just this `op`
colocation_groups = [(c, bind_op)
for (c, bind_op) in colocation_groups
if bind_op != op]
if colocation_groups:
device_to_bind_to = colocation_groups[-1][1].device
new_colocation_groups = [
c for (c, op) in colocation_groups
if op.device == device_to_bind_to
]
op._set_device(device_to_bind_to)
op._set_attr(
"_class",
pb2_AttrValue(list=pb2_AttrValue.ListValue(
s=new_colocation_groups)))
else:
try:
if op.get_attr("_class"):
op._clear_attr("_class")
except ValueError:
pass
def update_colocation_group(ops, old_op, new_op):
"""
For each op in ops, we replace the colocation group as old_op to colocation group as new_op.
Args:
ops (Iterable[Operation]): The operations to update
old_op (Operation): The op having the old colocation group
new_op (Operation): The op having the new colocation group
"""
old_groups = old_op.colocation_groups() or [
COLOCATION_PREFIX + as_bytes(new_op.name)
]
new_groups = new_op.colocation_groups() or [
COLOCATION_PREFIX + as_bytes(new_op.name)
]
for op in ops:
if op.colocation_groups() == old_groups:
op._set_attr("_class",
AttrValue(list=AttrValue.ListValue(s=new_groups)))
assert op.colocation_groups() == new_groups
def name_attr_list():
attr = {
'float': AttrValue(f=3.14159),
'list': AttrValue(list=AttrValue.ListValue(b=[True, False, True]))
}
return NameAttrList(name='test_name_attr_list', attr=attr)
def bool_list():
return AttrValue.ListValue(b=[True, False])
def int_list():
return AttrValue.ListValue(i=[1, 2, 3])
