def load(self, graph: Graph):
argv = graph.graph['cmd_params']
if argv.tensorflow_custom_layer_libraries:
libraries = argv.tensorflow_custom_layer_libraries.split(',')
for library in libraries:
log.info('Loading library "{}" with custom operations'.format(library))
tf_v1.load_op_library(library)
graph_def, variables_values = load_tf_graph_def(graph_file_name=argv.input_model,
is_binary=not argv.input_model_is_text,
checkpoint=argv.input_checkpoint,
user_output_node_names_list=argv.output,
model_dir=argv.saved_model_dir,
meta_graph_file=argv.input_meta_graph,
saved_model_tags=argv.saved_model_tags)
try:
tf_v1.import_graph_def(graph_def, name='')
except:
log.warning("TensorFlow post-processing of loaded model was unsuccessful. "
"This is an optional step that Model Optimizer performs for any input model but it is not usually "
"required for all models."
"It likely means that the original model is ill-formed. "
"Model Optimizer will continue converting this model.")
log.debug("Number of nodes in graph_def: {}".format(len(graph_def.node))) # pylint: disable=no-member
if argv.tensorboard_logdir:
tensorboard_util.dump_for_tensorboard(graph_def, argv.tensorboard_logdir)
update_extractors_with_extensions(tf_op_extractors)
try:
protobuf2nx(graph, graph_def)
except Exception as e:
raise Error(
'Cannot pre-process TensorFlow graph after reading from model file "{}". ' \
'File is corrupt or has unsupported format. Details: {}. ' +
refer_to_faq_msg(44),
argv.model_name,
str(e)
) from e
graph.__setattr__('name', argv.model_name)
# 'layout' parameter change may cause an issue in EltwiseInputReshape replacer
# and convert_nhwc_to_nchw(graph)
graph.graph['layout'] = 'NCHW' if argv.disable_nhwc_to_nchw else 'NHWC'
graph.graph['fw'] = 'tf'
graph.graph['variables_values'] = variables_values
del variables_values
restore_edges(graph, get_tf_edges)
remove_control_dependency_inputs(graph)
graph.check_empty_graph('protobuf2nx. It may happen due to problems with loaded model')
extract_node_attrs(graph, lambda node: tf_op_extractor(node, check_for_duplicates(tf_op_extractors)))
def _custom_cpp_op(op: CompilableOp, stateful, name):
""" Compiles and registers a custom C++ Tensorflow operator """
# Compile the .so file
tf_path = os.path.abspath(os.path.dirname(tf.__file__))
so_file = TFCompiler().compile_op(
op.name,
op.files,
op.inputs,
op.outputs,
any([f.endswith('.cu') for f in op.files]),
op.live_output,
additional_cmake_options=['-DTENSORFLOW_PATH=' + tf_path] +
op.cmake_options,
additional_definitions=op.defs,
output_folder=op.output_folder)
# Load the compiled library into Tensorflow
op_module = tf.load_op_library(so_file)
op_func = getattr(op_module, 'tf_op' + op.name)
op_grad_func = getattr(op_module, 'tf_op_grad' + op.name)
# Create the deep500 custom op object
lib = ctypes.CDLL(so_file)
if not getattr(lib, 'create_new_op', False):
raise ValueError('Invalid custom operator library file')
lib.create_new_op.restype = ctypes.c_int64
lib.is_cuda_supported.restype = ctypes.c_bool
lib.report.restype = ctypes.c_int64
return TFCompiledOp(op, op_func, op_grad_func, lib)
def testCustomOp(feedMat, corpus, chars, wordChars):
"decode using word beam search. Result is tuple, first entry is label string, second entry is char string."
# TF session
sess = tfv1.Session()
sess.run(tfv1.global_variables_initializer())
# load custom TF op
word_beam_search_module = tfv1.load_op_library(
'../cpp/proj/TFWordBeamSearch.so')
# input with shape TxBxC
mat = tfv1.placeholder(tf.float32, shape=feedMat.shape)
# decode using the "Words" mode of word beam search with beam width set to 25 and add-k smoothing to 0.0
assert len(chars) + 1 == mat.shape[2]
decode = word_beam_search_module.word_beam_search(mat, 25, 'Words', 0.0,
corpus.encode('utf8'),
chars.encode('utf8'),
wordChars.encode('utf8'))
# feed matrix of shape TxBxC and evaluate TF graph
res = sess.run(decode, {mat: feedMat})
# result is string of labels terminated by blank (similar to C-strings) if shorter than T
blank = len(chars)
s = ''
for label in res[0]:
if label == blank:
break
s += chars[label] # map label to char
return (res[0], s)
def setupCTC(self):
"create CTC loss and decoder and return them"
# BxTxC -> TxBxC
self.ctcIn3dTBC = tf.transpose(self.rnnOut3d, [1, 0, 2])
# ground truth text as sparse tensor
self.gtTexts = tf.SparseTensor(tf.placeholder(tf.int64, shape=[None, 2]) , tf.placeholder(tf.int32, [None]), tf.placeholder(tf.int64, [2]))
# calc loss for batch
self.seqLen = tf.placeholder(tf.int32, [None])
self.loss = tf.reduce_mean(tf.nn.ctc_loss(labels=self.gtTexts, inputs=self.ctcIn3dTBC, sequence_length=self.seqLen, ctc_merge_repeated=True))
# calc loss for each element to compute label probability
self.savedCtcInput = tf.placeholder(tf.float32, shape=[Model.maxTextLen, None, len(self.charList) + 1])
self.lossPerElement = tf.nn.ctc_loss(labels=self.gtTexts, inputs=self.savedCtcInput, sequence_length=self.seqLen, ctc_merge_repeated=True)
# decoder: either best path decoding or beam search decoding
if self.decoderType == DecoderType.BestPath:
self.decoder = tf.nn.ctc_greedy_decoder(inputs=self.ctcIn3dTBC, sequence_length=self.seqLen)
elif self.decoderType == DecoderType.BeamSearch:
self.decoder = tf.nn.ctc_beam_search_decoder(inputs=self.ctcIn3dTBC, sequence_length=self.seqLen, beam_width=50, merge_repeated=False)
elif self.decoderType == DecoderType.WordBeamSearch:
# import compiled word beam search operation (see https://github.com/githubharald/CTCWordBeamSearch)
word_beam_search_module = tf.load_op_library('TFWordBeamSearch.so')
# prepare information about language (dictionary, characters in dataset, characters forming words)
chars = str().join(self.charList)
wordChars = open('./model/model/wordCharList.txt').read().splitlines()[0]
corpus = open('./data/corpus.txt').read()
# decode using the "Words" mode of word beam search
self.decoder = word_beam_search_module.word_beam_search(tf.nn.softmax(self.ctcIn3dTBC, dim=2), 50, 'Words', 0.0, corpus.encode('utf8'), chars.encode('utf8'), wordChars.encode('utf8'))
def setupCTC(self):
""" Create CTC loss and decoder and return them """
# BxTxC -> TxBxC
self.ctcIn3dTBC = tf.transpose(a=self.rnnOut3d, perm=[1, 0, 2])
# Ground truth text as sparse tensor
with tf.compat.v1.name_scope('CTC_Loss'):
self.gtTexts = tf.SparseTensor(
tf.compat.v1.placeholder(tf.int64, shape=[None, 2]),
tf.compat.v1.placeholder(tf.int32, [None]),
tf.compat.v1.placeholder(tf.int64, [2]))
# Calculate loss for batch
self.seqLen = tf.compat.v1.placeholder(tf.int32, [None])
self.loss = tf.reduce_mean(input_tensor=tf.compat.v1.nn.ctc_loss(
labels=self.gtTexts,
inputs=self.ctcIn3dTBC,
sequence_length=self.seqLen,
ctc_merge_repeated=True,
ignore_longer_outputs_than_inputs=True))
with tf.compat.v1.name_scope('CTC_Decoder'):
# Decoder: Best path decoding or Word beam search decoding
if self.decoderType == DecoderType.BestPath:
self.decoder = tf.nn.ctc_greedy_decoder(
inputs=self.ctcIn3dTBC, sequence_length=self.seqLen)
elif self.decoderType == DecoderType.BeamSearch:
self.decoder = tf.compat.v1.nn.ctc_beam_search_decoder(
inputs=self.ctcIn3dTBC,
sequence_length=self.seqLen,
beam_width=50,
merge_repeated=True)
elif self.decoderType == DecoderType.WordBeamSearch:
# Import compiled word beam search operation (see https://github.com/githubharald/CTCWordBeamSearch)
word_beam_search_module = tf.load_op_library(
'./TFWordBeamSearch.so')
# Prepare: dictionary, characters in dataset, characters forming words
chars = codecs.open(FilePaths.wordCharList.txt, 'r').read()
wordChars = codecs.open(FilePaths.fnWordCharList, 'r').read()
corpus = codecs.open(FilePaths.corpus.txt, 'r').read()
# # Decoder using the "NGramsForecastAndSample": restrict number of (possible) next words to at most 20 words: O(W) mode of word beam search
# decoder = word_beam_search_module.word_beam_search(tf.nn.softmax(ctcIn3dTBC, dim=2), 25, 'NGramsForecastAndSample', 0.0, corpus.encode('utf8'), chars.encode('utf8'), wordChars.encode('utf8'))
# Decoder using the "Words": only use dictionary, no scoring: O(1) mode of word beam search
self.decoder = word_beam_search_module.word_beam_search(
tf.nn.softmax(self.ctcIn3dTBC, axis=2), 25, 'Words', 0.0,
corpus.encode('utf8'), chars.encode('utf8'),
wordChars.encode('utf8'))
# Return a CTC operation to compute the loss and CTC operation to decode the RNN output
return self.loss, self.decoder
def load_fl_ops_lib():
global _fl_ops_so
# Use double-checked locking to avoid taking lock unnecessarily.
if _fl_ops_so:
return _fl_ops_so
_fl_ops_so_lock.acquire()
try:
if _fl_ops_so:
return _fl_ops_so
_fl_ops_so = tf.load_op_library(dir_path + "/_fl_ops.so")
return _fl_ops_so
finally:
_fl_ops_so_lock.release()
def _loader_ops(so_path):
""" Post-building custom ops loading operations.
Args:
so_path Shared object path
"""
try:
lib = tf.load_op_library(str(so_path))
entries = lib.OP_LIST.ListFields()[0][1]
try:
while True:
opname = entries.pop().ListFields()[0][1]
opname = _camel_to_snake(opname)
register_op(opname, getattr(lib, opname))
except IndexError:
pass
except Exception as err:
with tools.Context(so_path.stem, "warning"):
print("Loading failed for custom op " + repr(str(so_path)) + ": " + str(err))
with tools.Context("traceback", "trace"):
traceback.print_exc()
def _processor(self, in_q, out_q, batch, rc):
tf.reset_default_graph()
processor_graph = tf.Graph()
with processor_graph.as_default() as g1:
with g1.name_scope("processor_graph") as scope:
op_module = tf.load_op_library('libkernel.so')
inp_batch = tf.placeholder(dtype=float,
shape=(batch, 3, 416, 416),
name="place_holder1")
out_s, out_m, out_l = op_module.My_Yolo_OP(
yolo_input=inp_batch)
init_op = tf.initialize_all_variables()
with tf.Session(graph=processor_graph) as sess:
sess.run(init_op)
while True:
curr_batch = in_q.get()
if curr_batch == 'STOP':
out_q.put('STOP')
break
yolo_out_s, yolo_out_m, yolo_out_l = sess.run(
[out_s, out_m, out_l], feed_dict={inp_batch: curr_batch})
rc.tick()
out_q.put((yolo_out_s, yolo_out_m, yolo_out_l))
''' Furthest point sampling
Original author: Haoqiang Fan
Modified by Charles R. Qi
All Rights Reserved. 2017.
'''
#import tensorflow as tf
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
from tensorflow.python.framework import ops
import sys
import os
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append(BASE_DIR)
sampling_module = tf.load_op_library(
os.path.join(BASE_DIR, 'tf_sampling_so.so'))
def prob_sample(inp, inpr):
'''
input:
batch_size * ncategory float32
batch_size * npoints float32
returns:
batch_size * npoints int32
'''
return sampling_module.prob_sample(inp, inpr)
ops.NoGradient('ProbSample')
import tensorflow.compat.v1 as tf
import os
#module_path = os.path.dirname(os.path.realpath(__file__))
#module_file = os.path.join(module_path, 'tf_mpi_ops.so')
#assert (os.path.isfile(module_file)), 'module tf_mpi_ops does not exist'
#mpi_ops = tf.load_op_library(module_file)
mpi_ops = tf.load_op_library('/home/zhaoxc/workspace/sw_hvd/sw_horovod/common/mpi_ops/mpi_ops/tf_mpi_ops.so')
def broadcast(in_tensor, root):
return mpi_ops.tf_broadcast(in_tensor, root=root)
def allreduce(in_tensor):
return mpi_ops.tf_allreduce(in_tensor)
def allgather(in_tensor, size):
return mpi_ops.tf_allgather(in_tensor, size=size)
def gather(in_tensor, root, rank, size):
return mpi_ops.tf_gather(in_tensor, root=root, rank=rank, size=size)
def allreduces(in_tensor_list, high_prec=0):
print("****************************now in allreduces*********************************\n")
print("about to get into mpi_ops.tf_allreduces\n")
return mpi_ops.tf_allreduces(in_tensor_list, precision=high_prec)
def broadcasts(in_tensor_list, root):
return mpi_ops.tf_broadcasts(in_tensor_list, root=root)
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
print(tf.__version__)
row_widths = tf.placeholder(dtype=tf.int32, shape=[None], name="width")
batch_size = tf.size(row_widths)
max_width = tf.reduce_max(row_widths)
indices = tf.to_int64(tf.where(tf.sequence_mask(row_widths, max_width)))
dense_shapes = tf.to_int64([batch_size, max_width])
with tf.Session() as sess:
x, y = sess.run([indices, dense_shapes],
feed_dict={"width:0": [2, 3, 4, 5]})
print(x)
print(x.shape)
print(y)
print("begin custom op")
sparse_module = tf.load_op_library('./sparse_transformer.so')
z = sparse_module.sparse_transformer(row_widths)
print("get the shape before run")
print(z.indices.get_shape())
print(z.shape.get_shape())
print("begin to run")
with tf.Session() as sess:
x, y = sess.run(z, feed_dict={"width:0": [2, 3, 4, 5]})
print(x.shape)
print(x)
print(y)
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Main code for creating the tree ensemble layer."""
import os
import tensorflow.compat.v1 as tf
from tensorflow import keras
from tensorflow.python.framework import ops
from tensorflow.keras.initializers import RandomUniform
# Assumes that neural_trees_ops.so is in tf_trees.
dir_path = os.path.dirname(os.path.realpath(__file__))
tf_trees_module = tf.load_op_library(dir_path + '/neural_trees_ops.so')
# Register the custom gradient for NTComputeOutputOp.
@ops.RegisterGradient('NTComputeOutputOp')
def _nt_compute_input_and_internal_params_gradients_op_cc(
op, grad_loss_wrt_tree_output, _):
"""Associated a custom gradient with an op."""
output_logits_dim = op.get_attr('output_logits_dim')
depth = op.get_attr('depth')
parallelize_over_samples = op.get_attr('parallelize_over_samples')
return [
tf_trees_module.nt_compute_input_and_internal_params_gradients_op(
grad_loss_wrt_tree_output, op.inputs[0], op.inputs[1], op.inputs[2],
op.inputs[3], output_logits_dim, depth, parallelize_over_samples),
None
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import candidate_sampling_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import embedding_ops
from tensorflow.python.ops import gen_array_ops
from tensorflow.python.ops import gen_nn_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import nn_ops
from tensorflow.python.ops import sparse_ops
from tensorflow.python.ops import variables
from tensorflow.python.util.deprecation import deprecated_args
from tensorflow.python.util.deprecation import deprecated_argument_lookup
from tensorflow.python.util.tf_export import tf_export
mpu_sim_conv2d_lib = tf.load_op_library(
'../../bin/build_mpusim_conv2d_release/mpusim-conv2d.so')
def mpusim_separable_conv2d_impl(input,
depthwise_filter,
pointwise_filter,
strides,
padding,
rate=None,
name=None,
data_format=None,
activations_datatype_size_byte=1,
weights_datatype_size_byte=1,
results_datatype_size_byte=4,
systolic_array_height=256,
systolic_array_width=256,
# Copyright 2018 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""When building as a dynamic .so provide a hook to load it at runtime."""
import tensorflow.compat.v1 as tf
# TODO: Store the returned object so application code can use it.
tf.load_op_library(
'multidim_image_augmentation/python/ops/_augmentation_ops.so')
log_device_placement=True)) as sess:
inputs_dict = {
self.input_tensor: self.input_data,
self.filter_tensor: self.filter_data
}
sess.run(tf.compat.v1.global_variables_initializer())
result = sess.run(
tf.abs(self.output_tensor - self.output_ref_tensor) /
tf.reduce_max(tf.abs(self.output_ref_tensor)),
feed_dict=inputs_dict)
print('{}: Linf Error: {}'.format(device, np.max(result)))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument(
'--device',
type=str,
help='Tensorflow device to be used (ie. \'cpu:0\', \'gpu:0\', ...).',
default='cpu:0')
args = parser.parse_args()
test_op_module = tf.load_op_library('libApproxGPUOpsTF.so')
test = TestApproxConv2D(test_op_module, [1, 256, 256, 1], [8, 8, 1, 1],
[1, 1, 1, 1])
test.run(args.device)
def get_module(module_name: str) -> "ModuleType":
"""Load force module.
Returns
-------
ModuleType
loaded force module
Raises
------
FileNotFoundError
if module is not found in directory
"""
if platform.system() == "Windows":
ext = ".dll"
#elif platform.system() == "Darwin":
# ext = ".dylib"
else:
ext = ".so"
module_file = ((Path(__file__).parent / SHARED_LIB_MODULE /
module_name).with_suffix(ext).resolve())
if not module_file.is_file():
raise FileNotFoundError(f"module {module_name} does not exist")
else:
try:
module = tf.load_op_library(str(module_file))
except tf.errors.NotFoundError as e:
# check CXX11_ABI_FLAG is compatiblity
# see https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html
# ABI should be the same
if 'CXX11_ABI_FLAG' in tf.__dict__:
tf_cxx11_abi_flag = tf.CXX11_ABI_FLAG
else:
tf_cxx11_abi_flag = tf.sysconfig.CXX11_ABI_FLAG
if TF_CXX11_ABI_FLAG != tf_cxx11_abi_flag:
raise RuntimeError(
"This deepmd-kit package was compiled with "
"CXX11_ABI_FLAG=%d, but TensorFlow runtime was compiled "
"with CXX11_ABI_FLAG=%d. These two library ABIs are "
"incompatible and thus an error is raised when loading %s. "
"You need to rebuild deepmd-kit against this TensorFlow "
"runtime." % (
TF_CXX11_ABI_FLAG,
tf_cxx11_abi_flag,
module_name,
)) from e
# different versions may cause incompatibility
# see #406, #447, #557, #774, and #796 for example
# throw a message if versions are different
if TF_VERSION != tf_py_version:
raise RuntimeError(
"The version of TensorFlow used to compile this "
"deepmd-kit package is %s, but the version of TensorFlow "
"runtime you are using is %s. These two versions are "
"incompatible and thus an error is raised when loading %s. "
"You need to install TensorFlow %s, or rebuild deepmd-kit "
"against TensorFlow %s.\nIf you are using a wheel from "
"pypi, you may consider to install deepmd-kit execuating "
"`pip install deepmd-kit --no-binary deepmd-kit` "
"instead." % (
TF_VERSION,
tf_py_version,
module_name,
TF_VERSION,
tf_py_version,
)) from e
raise RuntimeError(
"This deepmd-kit package is inconsitent with TensorFlow "
"Runtime, thus an error is raised when loading %s. "
"You need to rebuild deepmd-kit against this TensorFlow "
"runtime." % (module_name, )) from e
return module
from tensorflow.python import pywrap_tensorflow
import pkg_resources
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
coref_op_library = tf.load_op_library(
pkg_resources.resource_filename("e2edutch", "lib/coref_kernels.so"))
extract_spans = coref_op_library.extract_spans
tf.NotDifferentiable("ExtractSpans")
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os.path
# This is a placeholder for a Google-internal import.
import tensorflow.compat.v1 as tf
from tensorflow_serving.experimental.tensorflow.ops.remote_predict.ops import gen_remote_predict_op
# pylint: disable=wildcard-import
from tensorflow_serving.experimental.tensorflow.ops.remote_predict.ops.gen_remote_predict_op import *
# pylint: enable=wildcard-import
_remote_predict_op_module = tf.load_op_library(
os.path.join(tf.compat.v1.resource_loader.get_data_files_path(),
'_remote_predict_op.so'))
# Aliases
def run(input_tensor_alias,
input_tensors,
output_tensor_alias,
target_address,
model_name,
model_version=-1,
max_rpc_deadline_millis=3000,
output_types=None,
name=None,
signature_name='serving_default'):
"""Runs a predict in remote process through rpc.
import tensorflow.compat.v1 as tf
from tensorflow.python import pywrap_tensorflow
srl_op_library = tf.load_op_library("./srl_kernels.so")
extract_spans = srl_op_library.extract_spans
tf.no_gradient("ExtractSpans")
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Tests for ORQA ops."""
# from language.orqa import ops as orqa_ops
import tensorflow.compat.v1 as tf
orqa_ops = tf.load_op_library('exp/rlmfetops.so')
class OrqaOpsTest(tf.test.TestCase):
def test_reader_inputs(self):
concat_inputs = orqa_ops.reader_inputs(question_token_ids=[0, 1],
block_token_ids=[[2, 3, 4],
[5, 6, 0]],
block_lengths=[3, 2],
block_token_map=[[1, 2, 5],
[1, 3, 4]],
answer_token_ids=[[3, 4],
[7, 0]],
answer_lengths=[2, 1],
cls_token_id=10,
sep_token_id=11,
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
zero_out_module = tf.load_op_library('./zero_out.so')
x = zero_out_module.zero_out([[1, 2], [3, 4]])
with tf.Session() as sess:
y = sess.run(x)
print(x.shape)
print(y)
#import tensorflow as tf
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
from tensorflow.python.framework import ops
import sys
import os
BASE_DIR = os.path.dirname(__file__)
sys.path.append(BASE_DIR)
interpolate_module=tf.load_op_library(os.path.join(BASE_DIR, 'tf_interpolate_so.so'))
def three_nn(xyz1, xyz2):
'''
Input:
xyz1: (b,n,3) float32 array, unknown points
xyz2: (b,m,3) float32 array, known points
Output:
dist: (b,n,3) float32 array, distances to known points
idx: (b,n,3) int32 array, indices to known points
'''
return interpolate_module.three_nn(xyz1, xyz2)
ops.NoGradient('ThreeNN')
def three_interpolate(points, idx, weight):
'''
Input:
points: (b,m,c) float32 array, known points
idx: (b,n,3) int32 array, indices to known points
weight: (b,n,3) float32 array, weights on known points
Output:
out: (b,n,c) float32 array, interpolated point values
'''
return interpolate_module.three_interpolate(points, idx, weight)
@tf.RegisterGradient('ThreeInterpolate')
_HOME = os.path.abspath(os.path.join(os.path.dirname(__file__), '../../'))
_CPU_LIB_PATH = os.path.join(_HOME, 'cc/embedding.so')
class custom_fedlearner_operators_failed_to_load(object):
pass
lagrange_lite_ops = custom_fedlearner_operators_failed_to_load
path = _CPU_LIB_PATH
if os.path.exists(path):
lagrange_lite_ops = tf.load_op_library(path)
if lagrange_lite_ops is custom_fedlearner_operators_failed_to_load:
print("Failed to load fedlearner operators from %s" % path)
def _multidevice_preprocess_fids(fids, config, num_shards):
name = config['name']
with tf.name_scope("lagrange_multidevice_preprocess_fid/%s" % name):
slot_weight_index = tf.constant(config['slot_weight_index'],
dtype=tf.int64)
slot_hash_size = tf.constant(config['slot_hash_size'], dtype=tf.int64)
slot_weight_offset = tf.constant(config['slot_weight_offset'],
dtype=tf.int64)
ret = lagrange_lite_ops.lagrange_multi_device_preprocess_fid(
def driver(argv: argparse.Namespace):
"""
Convert TF GraphDef object to NetworkX representation.
The resulting graph is still TF-specific and needs normalization passes to be applied.
The specific TF structure assumes each GraphDef node is converted to a single
NetworkX node, node id is an original TF node name, and edges go directly from one op to another op.
"""
if argv.tensorflow_custom_layer_libraries:
libraries = argv.tensorflow_custom_layer_libraries.split(',')
for library in libraries:
log.info(
'Loading library "{}" with custom operations'.format(library))
tf_v1.load_op_library(library)
graph_def, variables_values = load_tf_graph_def(
graph_file_name=argv.input_model,
is_binary=not argv.input_model_is_text,
checkpoint=argv.input_checkpoint,
user_output_node_names_list=argv.output,
model_dir=argv.saved_model_dir,
meta_graph_file=argv.input_meta_graph,
saved_model_tags=argv.saved_model_tags)
try:
tf_v1.import_graph_def(graph_def, name='')
except:
log.warning(
"TensorFlow post-processing of loaded model was unsuccessful. "
"This is an optional step that Model Optimizer performs for any input model but it is not usually "
"required for all models."
"It likely means that the original model is ill-formed. "
"Model Optimizer will continue converting this model.")
log.debug("Number of nodes in graph_def: {}".format(len(graph_def.node))) # pylint: disable=no-member
if argv.tensorboard_logdir:
tensorboard_util.dump_for_tensorboard(graph_def,
argv.tensorboard_logdir)
update_extractors_with_extensions(tf_op_extractors)
try:
graph = protobuf2nx(graph_def)
graph.__setattr__('name', argv.model_name)
# 'layout' parameter change may cause an issue in EltwiseInputReshape replacer
# and convert_nhwc_to_nchw(graph)
graph.graph['layout'] = 'NCHW' if argv.disable_nhwc_to_nchw else 'NHWC'
graph.graph['cmd_params'] = argv
graph.graph['fw'] = 'tf'
graph.graph['ir_version'] = get_ir_version(argv)
graph.graph['variables_values'] = variables_values
del variables_values
graph = restore_edges(graph, get_tf_edges)
graph = remove_control_dependency_inputs(graph)
except Exception as e:
raise Error(
'Cannot pre-process TensorFlow graph after reading from model file "{}". ' \
'File is corrupt or has unsupported format. Details: {}. ' +
refer_to_faq_msg(44),
argv.model_name,
str(e)
) from e
graph.check_empty_graph(
'protobuf2nx. It may happen due to problems with loaded model')
extract_node_attrs(
graph, lambda node: tf_op_extractor(
node, check_for_duplicates(tf_op_extractors)))
# --------------------------------- LOAD END ------------------------------------------------------
class_registration.apply_replacements(graph, [
class_registration.ClassType.FRONT_REPLACER,
class_registration.ClassType.MIDDLE_REPLACER,
class_registration.ClassType.BACK_REPLACER
])
return graph
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
''' python custom ops '''
import tensorflow.compat.v1 as tf
from absl import logging
so_lib_file = tf.io.gfile.glob(tf.resource_loader.get_data_files_path() +
'/x_ops*.so')[0].split('/')[-1]
path = tf.resource_loader.get_path_to_datafile(so_lib_file)
logging.info('x_ops.so path:{}'.format(path))
gen_x_ops = tf.load_op_library(
tf.resource_loader.get_path_to_datafile(so_lib_file))
spectrum = gen_x_ops.spectrum
fbank = gen_x_ops.fbank
delta_delta = gen_x_ops.delta_delta
pitch = gen_x_ops.pitch
mfcc = gen_x_ops.mfcc_dct
frame_pow = gen_x_ops.frame_pow
speed = gen_x_ops.speed
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import tensorflow.compat.v1 as tf
from tensorflow.python.framework import ops
_op_library = tf.load_op_library(
os.path.join(os.path.dirname(__file__), 'tf_nndistance2.so'))
def nn_distance2(xyz1, xyz2):
"""
Computes the distance of nearest neighbors for a pair of point clouds.
Args:
xyz1: (batch_size, n1, 3) the first point cloud
xyz2: (batch_size, n2, 3) the second point cloud
Returns:
(dist1, idx1, dist2, idx2)
dist1: (batch_size, n1) squared distance from first to second
idx1: (batch_size, n1) nearest neighbor from first to second
dist2: (batch_size, n2) squared distance from second to first
idx2: (batch_size, n2) nearest neighbor from second to first
"""
return _op_library.nn_distance2(xyz1, xyz2)
@ops.RegisterGradient('NnDistance2')
def load(self, graph: Graph):
argv = graph.graph['cmd_params']
if argv.tensorflow_custom_layer_libraries:
libraries = argv.tensorflow_custom_layer_libraries.split(',')
for library in libraries:
log.info('Loading library "{}" with custom operations'.format(
library))
tf_v1.load_op_library(library)
graph_def, variables_values, framework, inputs_outputs_order = load_tf_graph_def(
graph_file_name=argv.input_model,
is_binary=not argv.input_model_is_text,
checkpoint=argv.input_checkpoint,
user_output_node_names_list=argv.output,
model_dir=argv.saved_model_dir,
meta_graph_file=argv.input_meta_graph,
saved_model_tags=argv.saved_model_tags)
if inputs_outputs_order is not None and isinstance(
inputs_outputs_order, tuple):
graph.inputs_order = inputs_outputs_order[0]
graph.outputs_order = inputs_outputs_order[1]
send_framework_info(framework)
try:
tf_v1.import_graph_def(graph_def, name='')
except:
log.warning(
"TensorFlow post-processing of loaded model was unsuccessful. "
"This is an optional step that Model Optimizer performs for any input model but it is not usually "
"required for all models. "
"It likely means that the original model is ill-formed. "
"Model Optimizer will continue converting this model.")
log.debug("Number of nodes in graph_def: {}".format(len(
graph_def.node))) # pylint: disable=no-member
if argv.tensorboard_logdir:
tensorboard_util.dump_for_tensorboard(graph_def,
argv.tensorboard_logdir)
update_extractors_with_extensions(tf_op_extractors)
try:
protobuf2nx(graph, graph_def)
except Exception as e:
raise Error(
'Cannot pre-process TensorFlow graph after reading from model file "{}". ' \
'File is corrupt or has unsupported format. Details: {}. ' +
refer_to_faq_msg(44),
argv.model_name,
str(e)
) from e
graph.__setattr__('name', argv.model_name)
# 'layout' parameter change may cause an issue in EltwiseInputReshape replacer
# and convert_nhwc_to_nchw(graph)
graph.graph['layout'] = 'NCHW' if argv.disable_nhwc_to_nchw else 'NHWC'
graph.graph['fw'] = 'tf'
graph.graph['variables_values'] = variables_values
del variables_values
used_tensors = restore_edges(graph, get_tf_edges)
# Tensor names information corresponding to a node is stored on outgoing edges.
# As output nodes do not have outgoing edges, fake outputs are required. In the following code
# for each output Identity node is added, and tensor name for the output is kept
# on (output, fake output) edge. After Result nodes adding transformation fake outputs
# are deleted from graph.
add_outputs_identity(
graph, graph.nodes - used_tensors,
lambda g, output, fake_node_name: g.add_edges_from(
[create_tf_edge(output, fake_node_name, 0)]))
remove_control_dependency_inputs(graph)
graph.check_empty_graph(
'protobuf2nx. It may happen due to problems with loaded model')
extract_node_attrs(
graph, lambda node: tf_op_extractor(
node, check_for_duplicates(tf_op_extractors)))
# try to detect layout from the nodes of the graph. If there are no convolution nodes in N(D)HWC layout then we
# consider that the graph is in NCHW layout and no layout conversion should be performed
if not argv.disable_nhwc_to_nchw and not graph_or_sub_graph_has_nhwc_ops(
graph):
if not argv.silent:
log.debug('disable_nhwc_to_nchw" was automatically enabled.')
for_graph_and_each_sub_graph_recursively(
graph, update_cmd_params_and_layout)
send_op_names_info(framework, graph)
send_shapes_info(framework, graph)
# coding=utf-8
# Copyright 2022 The Google Research Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Exposes C++ operations in Python."""
import tensorflow.compat.v1 as tf
try:
from sgk.sparse.ops.cc import gen_sparse_ops # pylint: disable=g-import-not-at-top
kernels = gen_sparse_ops
except ImportError:
# TODO(tgale): Avoid harcoding the library path.
kernels = tf.load_op_library("/usr/local/sgk/lib/libsgk.so")
def get_plugin(cuda_file, extra_nvcc_options=[]):
cuda_file_base = os.path.basename(cuda_file)
cuda_file_name, cuda_file_ext = os.path.splitext(cuda_file_base)
# Already in cache?
if cuda_file in _plugin_cache:
return _plugin_cache[cuda_file]
# Setup plugin.
if verbose:
print('Setting up TensorFlow plugin "%s": ' % cuda_file_base,
end='',
flush=True)
try:
# Hash CUDA source.
md5 = hashlib.md5()
with open(cuda_file, 'rb') as f:
md5.update(f.read())
md5.update(b'\n')
# Hash headers included by the CUDA code by running it through the preprocessor.
if not do_not_hash_included_headers:
if verbose:
print('Preprocessing... ', end='', flush=True)
with tempfile.TemporaryDirectory() as tmp_dir:
tmp_file = os.path.join(
tmp_dir, cuda_file_name + '_tmp' + cuda_file_ext)
_run_cmd(
_prepare_nvcc_cli(
'"%s" --preprocess -o "%s" --keep --keep-dir "%s"' %
(cuda_file, tmp_file, tmp_dir)))
with open(tmp_file, 'rb') as f:
bad_file_str = (
'"' + cuda_file.replace('\\', '/') + '"').encode(
'utf-8') # __FILE__ in error check macros
good_file_str = ('"' + cuda_file_base +
'"').encode('utf-8')
for ln in f:
if not ln.startswith(b'# ') and not ln.startswith(
b'#line '): # ignore line number pragmas
ln = ln.replace(bad_file_str, good_file_str)
md5.update(ln)
md5.update(b'\n')
# Select compiler options.
compile_opts = ''
if os.name == 'nt':
compile_opts += '"%s"' % os.path.join(
tf.sysconfig.get_lib(), 'python',
'_pywrap_tensorflow_internal.lib')
elif os.name == 'posix':
compile_opts += f' --compiler-options \'-fPIC\''
compile_opts += f' --compiler-options \'{" ".join(tf.sysconfig.get_compile_flags())}\''
compile_opts += f' --linker-options \'{" ".join(tf.sysconfig.get_link_flags())}\''
else:
assert False # not Windows or Linux, w00t?
compile_opts += f' --gpu-architecture={_get_cuda_gpu_arch_string()}'
compile_opts += ' --use_fast_math'
for opt in extra_nvcc_options:
compile_opts += ' ' + opt
nvcc_cmd = _prepare_nvcc_cli(compile_opts)
# Hash build configuration.
md5.update(('nvcc_cmd: ' + nvcc_cmd).encode('utf-8') + b'\n')
md5.update(('tf.VERSION: ' + tf.version.VERSION).encode('utf-8') +
b'\n')
md5.update(('cuda_cache_version_tag: ' +
cuda_cache_version_tag).encode('utf-8') + b'\n')
# Compile if not already compiled.
cache_dir = util.make_cache_dir_path(
'tflib-cudacache') if cuda_cache_path is None else cuda_cache_path
bin_file_ext = '.dll' if os.name == 'nt' else '.so'
bin_file = os.path.join(
cache_dir, cuda_file_name + '_' + md5.hexdigest() + bin_file_ext)
if not os.path.isfile(bin_file):
if verbose:
print('Compiling... ', end='', flush=True)
with tempfile.TemporaryDirectory() as tmp_dir:
tmp_file = os.path.join(tmp_dir,
cuda_file_name + '_tmp' + bin_file_ext)
_run_cmd(nvcc_cmd +
' "%s" --shared -o "%s" --keep --keep-dir "%s"' %
(cuda_file, tmp_file, tmp_dir))
os.makedirs(cache_dir, exist_ok=True)
intermediate_file = os.path.join(
cache_dir, cuda_file_name + '_' + uuid.uuid4().hex +
'_tmp' + bin_file_ext)
shutil.copyfile(tmp_file, intermediate_file)
os.rename(intermediate_file, bin_file) # atomic
# Load.
if verbose:
print('Loading... ', end='', flush=True)
plugin = tf.load_op_library(bin_file)
# Add to cache.
_plugin_cache[cuda_file] = plugin
if verbose:
print('Done.', flush=True)
return plugin
except:
if verbose:
print('Failed!', flush=True)
raise
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""ORQA ops."""
import os
import tensorflow.compat.v1 as tf
try:
orqa_ops
except NameError:
orqa_ops = tf.load_op_library(
os.path.join(os.path.dirname(os.path.abspath(__file__)), "orqa_ops.so"))
has_answer = orqa_ops.has_answer
reader_inputs = orqa_ops.reader_inputs
