def OptimizeGraph(rewriter_config, metagraph, graph_id=b'graph_to_optimize'):
"""Optimize the provided metagraph."""
with errors.raise_exception_on_not_ok_status() as status:
ret_from_swig = tf_opt.TF_OptimizeGraph(rewriter_config.SerializeToString(),
metagraph.SerializeToString(),
graph_id, status)
if ret_from_swig is None:
return None
out_graph = graph_pb2.GraphDef().FromString(ret_from_swig)
return out_graph
def OptimizeGraph(rewriter_config,
metagraph,
verbose=True,
graph_id=b'graph_to_optimize',
cluster=None):
"""Optimize the provided metagraph."""
with errors.raise_exception_on_not_ok_status() as status:
if cluster is None:
cluster = gcluster.Cluster()
ret_from_swig = tf_opt.TF_OptimizeGraph(
cluster.tf_cluster, rewriter_config.SerializeToString(),
metagraph.SerializeToString(), verbose, graph_id, status)
if ret_from_swig is None:
return None
out_graph = graph_pb2.GraphDef().FromString(ret_from_swig)
return out_graph
def OptimizeGraph(config_proto,
metagraph,
verbose=True,
graph_id=b'graph_to_optimize',
cluster=None):
"""Optimize the provided metagraph."""
if not isinstance(config_proto, config_pb2.ConfigProto):
raise TypeError('Expected config_proto to be a ConfigProto, saw type %s' %
type(config_proto))
if cluster is None:
cluster = gcluster.Cluster()
ret_from_swig = tf_opt.TF_OptimizeGraph(cluster.tf_cluster,
config_proto.SerializeToString(),
metagraph.SerializeToString(),
verbose, graph_id)
if ret_from_swig is None:
return None
out_graph = graph_pb2.GraphDef().FromString(ret_from_swig)
return out_graph
def OptimizeGraph(config_proto,
metagraph,
verbose=True,
graph_id=b'graph_to_optimize',
cluster=None,
strip_default_attributes=False):
"""Optimize the provided metagraph.
For best results, the signature_def field in `metagraph` should be populated
with information about input (feed) and output (fetch) tensors.
Args:
config_proto: a ConfigProto protobuf.
metagraph: a MetagraphDef protobuf.
verbose: whether to log optimization results.
graph_id: a string identifying this graph.
cluster: a grappler cluster object representing hardware resources
available to run this graph.
strip_default_attributes: whether graph node attributes having default
values should be removed after all the optimization passes. This
option is useful if the resulting graph will be executed by an older
process that might not know some of the recently added attributes.
"""
if not isinstance(config_proto, config_pb2.ConfigProto):
raise TypeError(
'Expected config_proto to be a ConfigProto, saw type %s' %
type(config_proto))
if cluster is None:
cluster = gcluster.Cluster()
ret_from_swig = tf_opt.TF_OptimizeGraph(cluster.tf_cluster,
config_proto.SerializeToString(),
metagraph.SerializeToString(),
verbose, graph_id,
strip_default_attributes)
if ret_from_swig is None:
return None
out_graph = graph_pb2.GraphDef().FromString(ret_from_swig)
return out_graph
def OptimizeGraph(config_proto,
metagraph,
verbose=True,
graph_id=b'graph_to_optimize',
cluster=None):
"""Optimize the provided metagraph.
For best results, the signature_def field in `metagraph` should be populated
with information about input (feed) and output (fetch) tensors.
"""
if not isinstance(config_proto, config_pb2.ConfigProto):
raise TypeError(
'Expected config_proto to be a ConfigProto, saw type %s' %
type(config_proto))
if cluster is None:
cluster = gcluster.Cluster()
ret_from_swig = tf_opt.TF_OptimizeGraph(cluster.tf_cluster,
config_proto.SerializeToString(),
metagraph.SerializeToString(),
verbose, graph_id)
if ret_from_swig is None:
return None
out_graph = graph_pb2.GraphDef().FromString(ret_from_swig)
return out_graph