def define_mnist_flags():
flags.DEFINE_integer('eval_secs', 10,
'How frequently to run evaluation step')
flags.DEFINE_integer('ckpt_steps', 100,
'How frequently to save a model checkpoin')
flags.DEFINE_integer('max_ckpts', 2,
'Maximum number of checkpoints to keep')
flags.DEFINE_integer('max_steps', os.environ.get('MAX_STEPS', 100),
'Max steps')
flags.DEFINE_integer('save_summary_steps', 10,
'How frequently to save TensorBoard summaries')
flags.DEFINE_integer('log_step_count_steps', 10,
'How frequently to log loss & global steps/s')
flags_core.define_base()
flags_core.define_performance(num_parallel_calls=False)
flags_core.define_image()
data_dir = os.path.abspath(
os.environ.get('PS_JOBSPACE', os.getcwd()) + '/data')
model_dir = os.path.abspath(
os.environ.get('PS_MODEL_PATH',
os.getcwd() + '/models') + '/mnist')
export_dir = os.path.abspath(
os.environ.get('PS_MODEL_PATH',
os.getcwd() + '/models'))
flags.adopt_module_key_flags(flags_core)
flags_core.set_defaults(
data_dir=data_dir,
model_dir=model_dir,
export_dir=export_dir,
train_epochs=int(os.environ.get('TRAIN_EPOCHS', 3)),
epochs_between_evals=int(os.environ.get('EPOCHS_EVAL', 5)),
batch_size=int(os.environ.get('BATCH_SIZE', 100)),
)
def define_common_bert_flags():
"""Define common flags for BERT tasks."""
flags_core.define_base(data_dir=False,
model_dir=True,
clean=False,
train_epochs=False,
epochs_between_evals=False,
stop_threshold=False,
batch_size=False,
num_gpu=True,
hooks=False,
export_dir=False,
distribution_strategy=True,
run_eagerly=True)
flags.DEFINE_string('bert_config_file', None,
'Bert configuration file to define core bert layers.')
flags.DEFINE_string(
'model_export_path', None,
'Path to the directory, where trainined model will be '
'exported.')
flags.DEFINE_string('tpu', '', 'TPU address to connect to.')
flags.DEFINE_string(
'init_checkpoint', None,
'Initial checkpoint (usually from a pre-trained BERT model).')
flags.DEFINE_integer('num_train_epochs', 3,
'Total number of training epochs to perform.')
flags.DEFINE_integer(
'steps_per_loop', 200,
'Number of steps per graph-mode loop. Only training step '
'happens inside the loop. Callbacks will not be called '
'inside.')
flags.DEFINE_float('learning_rate', 5e-5,
'The initial learning rate for Adam.')
flags.DEFINE_boolean(
'scale_loss', False,
'Whether to divide the loss by number of replica inside the per-replica '
'loss function.')
flags.DEFINE_boolean(
'use_keras_compile_fit', False,
'If True, uses Keras compile/fit() API for training logic. Otherwise '
'use custom training loop.')
flags.DEFINE_string(
'hub_module_url', None, 'TF-Hub path/url to Bert module. '
'If specified, init_checkpoint flag should not be used.')
# Adds flags for mixed precision training.
flags_core.define_performance(
num_parallel_calls=False,
inter_op=False,
intra_op=False,
synthetic_data=False,
max_train_steps=False,
dtype=True,
dynamic_loss_scale=True,
loss_scale=True,
all_reduce_alg=False,
num_packs=False,
enable_xla=True,
fp16_implementation=True,
)
def define_mnist_flags():
flags_core.define_base()
flags_core.define_performance(num_parallel_calls=False)
flags_core.define_image()
flags.adopt_module_key_flags(flags_core)
flags_core.set_defaults(data_dir='/tmp/mnist_data',
model_dir='/tmp/mnist_model',
batch_size=100,
train_epochs=40)
def define_mnist_flags():
flags_core.define_base()
flags_core.define_performance(num_parallel_calls=False)
flags_core.define_image()
flags.adopt_module_key_flags(flags_core)
flags_core.set_defaults(
data_dir=
"/workspace/zigangzhao/TensoFlowBDD/models/official/mnist/mnist_data",
model_dir=
"/workspace/zigangzhao/TensoFlowBDD/models/official/mnist/mnist_model/",
batch_size=100,
train_epochs=40)
def define_common_bert_flags():
"""Define the flags related to TPU's."""
flags.DEFINE_string('bert_config_file', None,
'Bert configuration file to define core bert layers.')
flags.DEFINE_string('model_dir', None, (
'The directory where the model weights and training/evaluation summaries '
'are stored. If not specified, save to /tmp/bert20/.'))
flags.DEFINE_string(
'model_export_path', None,
'Path to the directory, where trainined model will be '
'exported.')
flags.DEFINE_string('tpu', '', 'TPU address to connect to.')
flags.DEFINE_string(
'init_checkpoint', None,
'Initial checkpoint (usually from a pre-trained BERT model).')
flags.DEFINE_enum(
'strategy_type', 'mirror', ['tpu', 'mirror', 'multi_worker_mirror'],
'Distribution Strategy type to use for training. `tpu` uses '
'TPUStrategy for running on TPUs, `mirror` uses GPUs with single host, '
'`multi_worker_mirror` uses CPUs or GPUs with multiple hosts.')
flags.DEFINE_integer('num_train_epochs', 10,
'Total number of training epochs to perform.')
flags.DEFINE_integer(
'steps_per_loop', 100,
'Number of steps per graph-mode loop. Only training step '
'happens inside the loop. Callbacks will not be called '
'inside.')
flags.DEFINE_float('learning_rate', 5e-5,
'The initial learning rate for Adam.')
flags.DEFINE_boolean(
'run_eagerly', False,
'Run the model op by op without building a model function.')
flags.DEFINE_boolean(
'scale_loss', False,
'Whether to divide the loss by number of replica inside the per-replica '
'loss function.')
# Adds flags for mixed precision training.
flags_core.define_performance(
num_parallel_calls=False,
inter_op=False,
intra_op=False,
synthetic_data=False,
max_train_steps=False,
dtype=True,
dynamic_loss_scale=True,
loss_scale=True,
all_reduce_alg=False,
num_packs=False,
enable_xla=True,
fp16_implementation=True,
)
def define_mnist_flags():
flags_core.define_base()
flags_core.define_performance(num_parallel_calls=False)
flags_core.define_image()
data_dir = os.path.abspath(os.environ.get('PS_JOBSPACE', os.getcwd()) + '/data')
model_dir = os.path.abspath(os.environ.get('PS_MODEL_PATH', os.getcwd() + '/models') + '/mnist')
flags.adopt_module_key_flags(flags_core)
flags_core.set_defaults(data_dir=data_dir,
model_dir=model_dir,
export_dir=os.environ.get('PS_MODEL_PATH', os.getcwd() + '/models'),
batch_size=int(os.environ.get('batch_size', 100)),
epochs_between_evals=20,
train_epochs=int(os.environ.get('train_epochs', 40)))
def define_flags():
flags_core.define_base(clean=True,
num_gpu=False,
stop_threshold=True,
hooks=True,
train_epochs=True,
epochs_between_evals=True)
flags_core.define_performance(num_parallel_calls=True,
inter_op=True,
intra_op=True,
dynamic_loss_scale=True,
loss_scale=True,
synthetic_data=True,
dtype=True)
flags_core.define_image()
flags_core.define_benchmark()
def initialize_common_flags():
"""Define the common flags across models."""
define_common_hparams_flags()
flags_core.define_device(tpu=True)
flags_core.define_base(num_gpu=True,
model_dir=False,
data_dir=False,
batch_size=False)
flags_core.define_distribution(worker_hosts=True, task_index=True)
flags_core.define_performance(all_reduce_alg=True, num_packs=True)
# Reset the default value of num_gpus to zero.
FLAGS.num_gpus = 0
flags.DEFINE_string(
'strategy_type', 'mirrored', 'Type of distribute strategy.'
'One of mirrored, tpu and multiworker.')
def define_resnet_flags(resnet_size_choices=None):
"""Add flags and validators for ResNet."""
flags_core.define_base()
flags_core.define_performance(num_parallel_calls=False)
flags_core.define_image()
flags_core.define_benchmark()
flags.adopt_module_key_flags(flags_core)
flags.DEFINE_enum(
name='resnet_version',
short_name='rv',
default='1',
enum_values=['1', '2'],
help=flags_core.help_wrap(
'Version of ResNet. (1 or 2) See README.md for details.'))
flags.DEFINE_bool(
name='fine_tune',
short_name='ft',
default=False,
help=flags_core.help_wrap(
'If True do not train any parameters except for the final layer.'))
flags.DEFINE_string(
name='pretrained_model_checkpoint_path',
short_name='pmcp',
default=None,
help=flags_core.help_wrap(
'If not None initialize all the network except the final layer with '
'these values'))
flags.DEFINE_boolean(name='eval_only',
default=False,
help=flags_core.help_wrap(
'Skip training and only perform evaluation on '
'the latest checkpoint.'))
choice_kwargs = dict(
name='resnet_size',
short_name='rs',
default='50',
help=flags_core.help_wrap('The size of the ResNet model to use.'))
if resnet_size_choices is None:
flags.DEFINE_string(**choice_kwargs)
else:
flags.DEFINE_enum(enum_values=resnet_size_choices, **choice_kwargs)
def define_resnet_flags(resnet_size_choices=None):
"""Add flags and validators for ResNet."""
flags_core.define_base()
flags_core.define_performance(num_parallel_calls=False)
flags_core.define_image()
flags_core.define_benchmark()
flags.adopt_module_key_flags(flags_core)
flags.DEFINE_enum(
name='resnet_version',
short_name='rv',
default='2',
enum_values=['1', '2'],
help=flags_core.help_wrap(
'Version of ResNet. (1 or 2) See README.md for details.'))
choice_kwargs = dict(
name='resnet_size',
short_name='rs',
default='50',
help=flags_core.help_wrap('The size of the ResNet model to use.'))
if resnet_size_choices is None:
flags.DEFINE_string(**choice_kwargs)
else:
flags.DEFINE_enum(enum_values=resnet_size_choices, **choice_kwargs)
# The current implementation of ResNet v1 is numerically unstable when run
# with fp16 and will produce NaN errors soon after training begins.
msg = ('ResNet version 1 is not currently supported with fp16. '
'Please use version 2 instead.')
@flags.multi_flags_validator(['dtype', 'resnet_version'], message=msg)
def _forbid_v1_fp16(flag_values): # pylint: disable=unused-variable
return (flags_core.DTYPE_MAP[flag_values['dtype']][0] != tf.float16
or flag_values['resnet_version'] != '1')
def define_wide_deep_flags():
"""Add supervised learning flags, as well as wide-deep model type."""
flags_core.define_base()
flags_core.define_benchmark()
flags_core.define_performance(num_parallel_calls=False,
inter_op=True,
intra_op=True,
synthetic_data=False,
max_train_steps=False,
dtype=False,
all_reduce_alg=False)
flags.adopt_module_key_flags(flags_core)
flags.DEFINE_enum(name="model_type",
short_name="mt",
default="wide_deep",
enum_values=['wide', 'deep', 'wide_deep'],
help="Select model topology.")
flags.DEFINE_boolean(
name="download_if_missing",
default=True,
help=flags_core.help_wrap(
"Download data to data_dir if it is not already present."))
def define_transformer_flags():
"""Add flags and flag validators for running transformer_main."""
# Add common flags (data_dir, model_dir, train_epochs, etc.).
flags_core.define_base()
flags_core.define_performance(
num_parallel_calls=True,
inter_op=False,
intra_op=False,
synthetic_data=True,
max_train_steps=False,
dtype=False,
all_reduce_alg=True
)
flags_core.define_benchmark()
flags_core.define_device(tpu=True)
# Set flags from the flags_core module as "key flags" so they're listed when
# the '-h' flag is used. Without this line, the flags defined above are
# only shown in the full `--helpful` help text.
flags.adopt_module_key_flags(flags_core)
# Add transformer-specific flags
flags.DEFINE_enum(
name="param_set", short_name="mp", default="big",
enum_values=PARAMS_MAP.keys(),
help=flags_core.help_wrap(
"Parameter set to use when creating and training the model. The "
"parameters define the input shape (batch size and max length), "
"model configuration (size of embedding, # of hidden layers, etc.), "
"and various other settings. The big parameter set increases the "
"default batch size, embedding/hidden size, and filter size. For a "
"complete list of parameters, please see model/model_params.py."))
flags.DEFINE_bool(
name="static_batch", default=False,
help=flags_core.help_wrap(
"Whether the batches in the dataset should have static shapes. In "
"general, this setting should be False. Dynamic shapes allow the "
"inputs to be grouped so that the number of padding tokens is "
"minimized, and helps model training. In cases where the input shape "
"must be static (e.g. running on TPU), this setting will be ignored "
"and static batching will always be used."))
# Flags for training with steps (may be used for debugging)
flags.DEFINE_integer(
name="train_steps", short_name="ts", default=None,
help=flags_core.help_wrap("The number of steps used to train."))
flags.DEFINE_integer(
name="steps_between_evals", short_name="sbe", default=1000,
help=flags_core.help_wrap(
"The Number of training steps to run between evaluations. This is "
"used if --train_steps is defined."))
# BLEU score computation
flags.DEFINE_string(
name="bleu_source", short_name="bls", default=None,
help=flags_core.help_wrap(
"Path to source file containing text translate when calculating the "
"official BLEU score. Both --bleu_source and --bleu_ref must be set. "
"Use the flag --stop_threshold to stop the script based on the "
"uncased BLEU score."))
flags.DEFINE_string(
name="bleu_ref", short_name="blr", default=None,
help=flags_core.help_wrap(
"Path to source file containing text translate when calculating the "
"official BLEU score. Both --bleu_source and --bleu_ref must be set. "
"Use the flag --stop_threshold to stop the script based on the "
"uncased BLEU score."))
flags.DEFINE_string(
name="vocab_file", short_name="vf", default=None,
help=flags_core.help_wrap(
"Path to subtoken vocabulary file. If data_download.py was used to "
"download and encode the training data, look in the data_dir to find "
"the vocab file."))
flags_core.set_defaults(data_dir="/tmp/translate_ende",
model_dir="/tmp/transformer_model",
batch_size=None,
train_epochs=None)
@flags.multi_flags_validator(
["train_epochs", "train_steps"],
message="Both --train_steps and --train_epochs were set. Only one may be "
"defined.")
def _check_train_limits(flag_dict):
return flag_dict["train_epochs"] is None or flag_dict["train_steps"] is None
@flags.multi_flags_validator(
["bleu_source", "bleu_ref"],
message="Both or neither --bleu_source and --bleu_ref must be defined.")
def _check_bleu_files(flags_dict):
return (flags_dict["bleu_source"] is None) == (
flags_dict["bleu_ref"] is None)
@flags.multi_flags_validator(
["bleu_source", "bleu_ref", "vocab_file"],
message="--vocab_file must be defined if --bleu_source and --bleu_ref "
"are defined.")
def _check_bleu_vocab_file(flags_dict):
if flags_dict["bleu_source"] and flags_dict["bleu_ref"]:
return flags_dict["vocab_file"] is not None
return True
@flags.multi_flags_validator(
["export_dir", "vocab_file"],
message="--vocab_file must be defined if --export_dir is set.")
def _check_export_vocab_file(flags_dict):
if flags_dict["export_dir"]:
return flags_dict["vocab_file"] is not None
return True
flags_core.require_cloud_storage(["data_dir", "model_dir", "export_dir"])
def define_flags():
flags_core.define_base(num_gpu=False)
flags_core.define_performance(dynamic_loss_scale=True, loss_scale=True)
flags_core.define_image()
flags_core.define_benchmark()
def define_keras_flags(dynamic_loss_scale=True):
"""Define flags for Keras models."""
flags_core.define_base(clean=True,
num_gpu=True,
run_eagerly=True,
train_epochs=True,
epochs_between_evals=True,
distribution_strategy=True)
flags_core.define_performance(num_parallel_calls=False,
synthetic_data=True,
dtype=True,
all_reduce_alg=True,
num_packs=True,
tf_gpu_thread_mode=True,
datasets_num_private_threads=True,
dynamic_loss_scale=dynamic_loss_scale,
loss_scale=True,
fp16_implementation=True,
tf_data_experimental_slack=True,
enable_xla=True,
force_v2_in_keras_compile=True,
training_dataset_cache=True)
flags_core.define_image()
flags_core.define_benchmark()
flags_core.define_distribution()
flags.adopt_module_key_flags(flags_core)
flags.DEFINE_boolean(name='enable_eager',
default=False,
help='Enable eager?')
flags.DEFINE_boolean(name='skip_eval',
default=False,
help='Skip evaluation?')
# TODO(b/135607288): Remove this flag once we understand the root cause of
# slowdown when setting the learning phase in Keras backend.
flags.DEFINE_boolean(
name='set_learning_phase_to_train',
default=True,
help='If skip eval, also set Keras learning phase to 1 (training).')
flags.DEFINE_boolean(
name='explicit_gpu_placement',
default=False,
help='If not using distribution strategy, explicitly set device scope '
'for the Keras training loop.')
flags.DEFINE_boolean(name='use_trivial_model',
default=False,
help='Whether to use a trivial Keras model.')
flags.DEFINE_boolean(name='report_accuracy_metrics',
default=True,
help='Report metrics during training and evaluation.')
flags.DEFINE_boolean(
name='use_tensor_lr',
default=False,
help='Use learning rate tensor instead of a callback.')
flags.DEFINE_boolean(name='enable_tensorboard',
default=False,
help='Whether to enable Tensorboard callback.')
flags.DEFINE_integer(
name='train_steps',
default=None,
help='The number of steps to run for training. If it is larger than '
'# batches per epoch, then use # batches per epoch. This flag will be '
'ignored if train_epochs is set to be larger than 1. ')
flags.DEFINE_string(
name='profile_steps',
default=None,
help=
'Save profiling data to model dir at given range of global steps. The '
'value must be a comma separated pair of positive integers, specifying '
'the first and last step to profile. For example, "--profile_steps=2,4" '
'triggers the profiler to process 3 steps, starting from the 2nd step. '
'Note that profiler has a non-trivial performance overhead, and the '
'output file can be gigantic if profiling many steps.')
flags.DEFINE_boolean(
name='batchnorm_spatial_persistent',
default=True,
help='Enable the spacial persistent mode for CuDNN batch norm kernel.')
flags.DEFINE_boolean(
name='enable_get_next_as_optional',
default=False,
help='Enable get_next_as_optional behavior in DistributedIterator.')
flags.DEFINE_boolean(
name='enable_checkpoint_and_export',
default=False,
help=
'Whether to enable a checkpoint callback and export the savedmodel.')
flags.DEFINE_string(name='tpu',
default='',
help='TPU address to connect to.')
flags.DEFINE_integer(
name='steps_per_loop',
default=1,
help='Number of steps per graph-mode loop. Only training step happens '
'inside the loop. Callbacks will not be called inside. Will be capped at '
'steps per epoch.')
def define_resnet_flags(resnet_size_choices=None):
"""Add flags and validators for ResNet."""
flags_core.define_base()
flags_core.define_performance(num_parallel_calls=False,
tf_gpu_thread_mode=True,
datasets_num_private_threads=True,
datasets_num_parallel_batches=True)
flags_core.define_image()
flags_core.define_benchmark()
flags.adopt_module_key_flags(flags_core)
flags.DEFINE_enum(
name='resnet_version',
short_name='rv',
default='2',
enum_values=['1', '2'],
help=flags_core.help_wrap(
'Version of ResNet. (1 or 2) See README.md for details.'))
flags.DEFINE_bool(
name='fine_tune',
short_name='ft',
default=False,
help=flags_core.help_wrap(
'If True do not train any parameters except for the final layer.'))
flags.DEFINE_string(
name='pretrained_model_checkpoint_path',
short_name='pmcp',
default=None,
help=flags_core.help_wrap(
'If not None initialize all the network except the final layer with '
'these values'))
flags.DEFINE_boolean(name='eval_only',
default=False,
help=flags_core.help_wrap(
'Skip training and only perform evaluation on '
'the latest checkpoint.'))
flags.DEFINE_boolean(
name='image_bytes_as_serving_input',
default=False,
help=flags_core.help_wrap(
'If True exports savedmodel with serving signature that accepts '
'JPEG image bytes instead of a fixed size [HxWxC] tensor that '
'represents the image. The former is easier to use for serving at '
'the expense of image resize/cropping being done as part of model '
'inference. Note, this flag only applies to ImageNet and cannot '
'be used for CIFAR.'))
flags.DEFINE_float(name='reconst_loss_scale',
default=10.0,
help=flags_core.help_wrap('scale the reconst_loss'))
flags.DEFINE_boolean(
name='use_ce',
default=False,
help=flags_core.help_wrap(
'use cross entropy loss for compressive sensing training'))
flags.DEFINE_string(
name='optimizer',
short_name='opt',
# default='sgd',
default='adam',
help=flags_core.help_wrap('Choose optimizer for training'))
flags.DEFINE_boolean(
name='clip_grad',
default=False,
help=flags_core.help_wrap('whether to clip weights during training'))
flags.DEFINE_boolean(name='spectral_norm',
short_name='sn',
default=True,
help=flags_core.help_wrap(
'whether to user spectral norm in the cs part'))
flags.DEFINE_float(name='ce_scale',
default=1.0,
help=flags_core.help_wrap('scale the cross_entropy'))
flags.DEFINE_boolean(
name='sep_grad_nrom',
default=False,
help=flags_core.help_wrap(
'spearate the gradients from reconstruction and ce, and norm the ce grad'
))
flags.DEFINE_boolean(
name='norm_teach_feature',
default=False,
help=flags_core.help_wrap(
'norm each channel of teaching feature with BN params'))
flags.DEFINE_boolean(name='no_dense_init',
default=False,
help=flags_core.help_wrap(
'dont init resenet/dense during fine tuning'))
flags.DEFINE_float(name='compress_ratio',
default=0.1,
help=flags_core.help_wrap(
'the compress ratio of the offloading layer'))
choice_kwargs = dict(
name='resnet_size',
short_name='rs',
default='50',
help=flags_core.help_wrap('The size of the ResNet model to use.'))
if resnet_size_choices is None:
flags.DEFINE_string(**choice_kwargs)
else:
flags.DEFINE_enum(enum_values=resnet_size_choices, **choice_kwargs)
def define_keras_benchmark_flags():
"""Add flags for keras built-in application models."""
flags_core.define_base(hooks=False)
flags_core.define_performance()
flags_core.define_image()
flags_core.define_benchmark()
flags.adopt_module_key_flags(flags_core)
flags_core.set_defaults(data_format="channels_last",
use_synthetic_data=True,
batch_size=32,
train_epochs=2)
flags.DEFINE_enum(name="model",
default=None,
enum_values=MODELS.keys(),
case_sensitive=False,
help=flags_core.help_wrap("Model to be benchmarked."))
flags.DEFINE_integer(
name="num_train_images",
default=1000,
help=flags_core.help_wrap(
"The number of synthetic images for training. The default value is "
"1000."))
flags.DEFINE_integer(
name="num_eval_images",
default=50,
help=flags_core.help_wrap(
"The number of synthetic images for evaluation. The default value is "
"50."))
flags.DEFINE_boolean(
name="eager",
default=False,
help=flags_core.help_wrap(
"To enable eager execution. Note that if eager execution is enabled, "
"only one GPU is utilized even if multiple GPUs are provided and "
"multi_gpu_model is used."))
flags.DEFINE_boolean(
name="dist_strat",
default=False,
help=flags_core.help_wrap(
"To enable distribution strategy for model training and evaluation. "
"Number of GPUs used for distribution strategy can be set by the "
"argument --num_gpus."))
flags.DEFINE_list(
name="callbacks",
default=["ExamplesPerSecondCallback", "LoggingMetricCallback"],
help=flags_core.help_wrap(
"A list of (case insensitive) strings to specify the names of "
"callbacks. For example: `--callbacks ExamplesPerSecondCallback,"
"LoggingMetricCallback`"))
@flags.multi_flags_validator(
["eager", "dist_strat"],
message="Both --eager and --dist_strat were set. Only one can be "
"defined, as DistributionStrategy is not supported in Eager "
"execution currently.")
# pylint: disable=unused-variable
def _check_eager_dist_strat(flag_dict):
return not (flag_dict["eager"] and flag_dict["dist_strat"])
def define_ncf_flags():
"""Add flags for running ncf_main."""
# Add common flags
flags_core.define_base(model_dir=True, clean=True, train_epochs=True,
epochs_between_evals=True, export_dir=False,
run_eagerly=True, stop_threshold=True, num_gpu=True,
hooks=True, distribution_strategy=True)
flags_core.define_performance(
synthetic_data=True,
dtype=True,
fp16_implementation=True,
loss_scale=True,
dynamic_loss_scale=True,
enable_xla=True,
force_v2_in_keras_compile=True
)
flags_core.define_device(tpu=True)
flags_core.define_benchmark()
flags.adopt_module_key_flags(flags_core)
flags_core.set_defaults(
model_dir="/tmp/ncf/",
data_dir="/tmp/movielens-data/",
train_epochs=2,
batch_size=256,
hooks="ProfilerHook",
tpu=None
)
# Add ncf-specific flags
flags.DEFINE_enum(
name="dataset", default="ml-1m",
enum_values=["ml-1m", "ml-20m", "ml-20mx16x32"], case_sensitive=False,
help=flags_core.help_wrap(
"Dataset to be trained and evaluated."))
flags.DEFINE_boolean(
name="download_if_missing", default=True, help=flags_core.help_wrap(
"Download data to data_dir if it is not already present."))
flags.DEFINE_integer(
name="eval_batch_size", default=None, help=flags_core.help_wrap(
"The batch size used for evaluation. This should generally be larger"
"than the training batch size as the lack of back propagation during"
"evaluation can allow for larger batch sizes to fit in memory. If not"
"specified, the training batch size (--batch_size) will be used."))
flags.DEFINE_integer(
name="num_factors", default=8,
help=flags_core.help_wrap("The Embedding size of MF model."))
# Set the default as a list of strings to be consistent with input
# arguments
flags.DEFINE_list(
name="layers", default=["64", "32", "16", "8"],
help=flags_core.help_wrap(
"The sizes of hidden layers for MLP. Example "
"to specify different sizes of MLP layers: --layers=32,16,8,4"))
flags.DEFINE_float(
name="mf_regularization", default=0., help=flags_core.help_wrap(
"The regularization factor for MF embeddings. The factor is used by "
"regularizer which allows to apply penalties on layer parameters or "
"layer activity during optimization."))
flags.DEFINE_list(
name="mlp_regularization", default=[
"0.", "0.", "0.", "0."], help=flags_core.help_wrap(
"The regularization factor for each MLP layer. See mf_regularization "
"help for more info about regularization factor."))
flags.DEFINE_integer(name="num_neg", default=4, help=flags_core.help_wrap(
"The Number of negative instances to pair with a positive instance."))
flags.DEFINE_float(
name="learning_rate", default=0.001,
help=flags_core.help_wrap("The learning rate."))
flags.DEFINE_float(name="beta1", default=0.9, help=flags_core.help_wrap(
"beta1 hyperparameter for the Adam optimizer."))
flags.DEFINE_float(name="beta2", default=0.999, help=flags_core.help_wrap(
"beta2 hyperparameter for the Adam optimizer."))
flags.DEFINE_float(
name="epsilon", default=1e-8,
help=flags_core.help_wrap("epsilon hyperparameter for the Adam "
"optimizer."))
flags.DEFINE_float(
name="hr_threshold", default=1.0,
help=flags_core.help_wrap(
"If passed, training will stop when the evaluation metric HR is "
"greater than or equal to hr_threshold. For dataset ml-1m, the "
"desired hr_threshold is 0.68 which is the result from the paper; "
"For dataset ml-20m, the threshold can be set as 0.95 which is "
"achieved by MLPerf implementation."))
flags.DEFINE_enum(
name="constructor_type", default="bisection",
enum_values=["bisection", "materialized"], case_sensitive=False,
help=flags_core.help_wrap(
"Strategy to use for generating false negatives. materialized has a"
"precompute that scales badly, but a faster per-epoch construction"
"time and can be faster on very large systems."))
flags.DEFINE_string(
name="train_dataset_path",
default=None,
help=flags_core.help_wrap("Path to training data."))
flags.DEFINE_string(
name="eval_dataset_path",
default=None,
help=flags_core.help_wrap("Path to evaluation data."))
flags.DEFINE_string(
name="input_meta_data_path",
default=None,
help=flags_core.help_wrap("Path to input meta data file."))
flags.DEFINE_bool(
name="ml_perf", default=False, help=flags_core.help_wrap(
"If set, changes the behavior of the model slightly to match the "
"MLPerf reference implementations here: \n"
"https://github.com/mlperf/reference/tree/master/recommendation/"
"pytorch\n"
"The two changes are:\n"
"1. When computing the HR and NDCG during evaluation, remove "
"duplicate user-item pairs before the computation. This results in "
"better HRs and NDCGs.\n"
"2. Use a different soring algorithm when sorting the input data, "
"which performs better due to the fact the sorting algorithms are "
"not stable."))
flags.DEFINE_bool(
name="output_ml_perf_compliance_logging",
default=False,
help=flags_core.help_wrap(
"If set, output the MLPerf compliance logging. This is only useful "
"if one is running the model for MLPerf. See "
"https://github.com/mlperf/policies/blob/master/training_rules.adoc"
"#submission-compliance-logs for details. This uses sudo and so may "
"ask for your password, as root access is needed to clear the system "
"caches, which is required for MLPerf compliance."))
flags.DEFINE_integer(
name="seed", default=None, help=flags_core.help_wrap(
"This value will be used to seed both NumPy and TensorFlow."))
@flags.validator("eval_batch_size", "eval_batch_size must be at least {}"
.format(rconst.NUM_EVAL_NEGATIVES + 1))
def eval_size_check(eval_batch_size):
return (eval_batch_size is None or
int(eval_batch_size) > rconst.NUM_EVAL_NEGATIVES)
flags.DEFINE_bool(
name="use_xla_for_gpu", default=False, help=flags_core.help_wrap(
"If True, use XLA for the model function. Only works when using a "
"GPU. On TPUs, XLA is always used"))
xla_message = "--use_xla_for_gpu is incompatible with --tpu"
@flags.multi_flags_validator(
["use_xla_for_gpu", "tpu"], message=xla_message)
def xla_validator(flag_dict):
return not flag_dict["use_xla_for_gpu"] or not flag_dict["tpu"]
flags.DEFINE_bool(
name="early_stopping",
default=False,
help=flags_core.help_wrap(
"If True, we stop the training when it reaches hr_threshold"))
flags.DEFINE_bool(
name="keras_use_ctl",
default=False,
help=flags_core.help_wrap(
"If True, we use a custom training loop for keras."))
def define_resnet_flags(resnet_size_choices=None,
dynamic_loss_scale=False,
fp16_implementation=False):
"""Add flags and validators for ResNet."""
flags_core.define_base()
flags_core.define_performance(num_parallel_calls=False,
tf_gpu_thread_mode=True,
datasets_num_private_threads=True,
dynamic_loss_scale=dynamic_loss_scale,
fp16_implementation=fp16_implementation,
loss_scale=True,
tf_data_experimental_slack=True)
flags_core.define_image()
flags_core.define_benchmark()
flags.adopt_module_key_flags(flags_core)
flags.DEFINE_enum(
name='resnet_version',
short_name='rv',
default='2',
enum_values=['1', '2'],
help=flags_core.help_wrap(
'Version of ResNet. (1 or 2) See README.md for details.'))
flags.DEFINE_bool(
name='fine_tune',
short_name='ft',
default=False,
help=flags_core.help_wrap(
'If True do not train any parameters except for the final layer.'))
flags.DEFINE_string( # "/home/zxc/Liu/models-master-new/official/r1/resnet/model/"
name='pretrained_model_checkpoint_path',
short_name='pmcp',
default="/home/zxc/Liu/models-master-new/official/r1/resnet/model/",
# default=None,
help=flags_core.help_wrap(
'If not None initialize all the network except the final layer with '
'these values'))
flags.DEFINE_boolean(name='eval_only',
default=False,
help=flags_core.help_wrap(
'Skip training and only perform evaluation on '
'the latest checkpoint.'))
flags.DEFINE_boolean(
name='image_bytes_as_serving_input',
default=False,
help=flags_core.help_wrap(
'If True exports savedmodel with serving signature that accepts '
'JPEG image bytes instead of a fixed size [HxWxC] tensor that '
'represents the image. The former is easier to use for serving at '
'the expense of image resize/cropping being done as part of model '
'inference. Note, this flag only applies to ImageNet and cannot '
'be used for CIFAR.'))
flags.DEFINE_boolean(
name='use_train_and_evaluate',
default=False,
help=flags_core.help_wrap(
'If True, uses `tf.estimator.train_and_evaluate` for the training '
'and evaluation loop, instead of separate calls to `classifier.train '
'and `classifier.evaluate`, which is the default behavior.'))
flags.DEFINE_string(
name='worker_hosts',
default=None,
help=flags_core.help_wrap(
'Comma-separated list of worker ip:port pairs for running '
'multi-worker models with DistributionStrategy. The user would '
'start the program on each host with identical value for this flag.'
))
flags.DEFINE_integer(name='task_index',
default=-1,
help=flags_core.help_wrap(
'If multi-worker training, the task_index of '
'this worker.'))
flags.DEFINE_bool(name='enable_lars',
default=False,
help=flags_core.help_wrap(
'Enable LARS optimizer for large batch training.'))
flags.DEFINE_float(
name='label_smoothing',
default=0.0,
help=flags_core.help_wrap(
'Label smoothing parameter used in the softmax_cross_entropy'))
flags.DEFINE_float(name='weight_decay',
default=1e-4,
help=flags_core.help_wrap(
'Weight decay coefficiant for l2 regularization.'))
choice_kwargs = dict(
name='resnet_size',
short_name='rs',
default='50',
help=flags_core.help_wrap('The size of the ResNet model to use.'))
if resnet_size_choices is None:
flags.DEFINE_string(**choice_kwargs)
else:
flags.DEFINE_enum(enum_values=resnet_size_choices, **choice_kwargs)
def define_flags():
flags_core.define_base(num_gpu=False)
flags_core.define_performance()
flags_core.define_image()
flags_core.define_benchmark()
