Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add distribution_strategy and all_reduce_alg flags to TensorFlow BERT pretraining #745

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
Open

Add distribution_strategy and all_reduce_alg flags to TensorFlow BERT pretraining #745

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
16ca431
Add `distribution_strategy` and `all_reduce_alg` flags to TensorFlow …
rapsealk Jun 5, 2024
77934d0
fix: Use 'FLAGS' instead of 'flags'
rapsealk Jun 5, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 24 additions & 2 deletions language_model/tensorflow/bert/run_pretraining.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,6 +117,28 @@
flags.DEFINE_integer("keep_checkpoint_max", 5,
"The maximum number of checkpoints to keep.")

flags.DEFINE_string(
name="distribution_strategy", short_name="ds", default="mirrored",
help="The Distribution Strategy to use for training. "
"Accepted values are 'off', 'one_device', "
"'mirrored', 'parameter_server', 'collective', "
"case insensitive. 'off' means not to use "
"Distribution Strategy; 'default' means to choose "
"from `MirroredStrategy` or `OneDeviceStrategy` "
"according to the number of GPUs.")

flags.DEFINE_string(
name="all_reduce_alg", short_name="ara", default="nccl",
help="Defines the algorithm to use for performing all-reduce."
"When specified with MirroredStrategy for single "
"worker, this controls "
"tf.contrib.distribute.AllReduceCrossTowerOps. When "
"specified with MultiWorkerMirroredStrategy, this "
"controls "
"tf.distribute.experimental.CollectiveCommunication; "
"valid options are `ring` and `nccl`.")


def model_fn_builder(bert_config, init_checkpoint, learning_rate,
num_train_steps, num_warmup_steps, use_tpu,
use_one_hot_embeddings, optimizer, poly_power,
Expand Down Expand Up @@ -542,9 +564,9 @@ def main(_):
allow_soft_placement=True)

distribution_strategy = distribution_utils.get_distribution_strategy(
distribution_strategy="mirrored",
distribution_strategy=FLAGS.distribution_strategy,
num_gpus=FLAGS.num_gpus,
all_reduce_alg="nccl",
all_reduce_alg=FLAGS.all_reduce_alg,
num_packs=0)

dist_gpu_config = tf.estimator.RunConfig(
Expand Down
Loading