Skip to content

[Bug Fix] Make DDP workers set absl logging as default #1824

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

Merged
QuantuMope merged 2 commits into from
Dec 10, 2025
Merged

[Bug Fix] Make DDP workers set absl logging as default #1824

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b21f7c9
[Bug Fix] Make DDP workers set absl logging as default
QuantuMope Dec 10, 2025
ff8b3f8
Remove program name as it creates duplicate logs
QuantuMope Dec 10, 2025
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
11 changes: 7 additions & 4 deletions alf/bin/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -118,16 +118,19 @@ def check_valid_launch():
assert not extra, f"Unexpected environment variables for non-distributed launch: {extra}"


def _setup_logging(rank: int, log_dir: str):
def _setup_logging(log_dir: str):
"""Setup logging for each process

Args:
rank (int): The ID of the process among all of the DDP processes
log_dir (str): path to the directory where log files are written to
"""
FLAGS.alsologtostderr = True
logging.set_verbosity(logging.INFO)
logging.get_absl_handler().use_absl_log_file(log_dir=log_dir)
# Spawned subprocesses create a new interpreter so will change the
# default logging back to python's logging module.
# For DDP worker logging to work, we need to explicitly set it back to absl.
logging.use_absl_handler()


def _setup_device():
Expand Down Expand Up @@ -232,7 +235,7 @@ def training_worker(rank: int,
in different worker processes, if multi-gpu training is used.
"""
try:
_setup_logging(log_dir=root_dir, rank=rank)
_setup_logging(log_dir=root_dir)
_setup_device()
if world_size > 1:
# Specialization for distributed mode
Expand Down Expand Up @@ -298,7 +301,7 @@ def training_worker_multi_node(local_rank: int,
in different worker processes, if multi-gpu training is used.
"""
try:
_setup_logging(log_dir=root_dir, rank=rank)
_setup_logging(log_dir=root_dir)
_setup_device()

# Specialization for distributed mode
Expand Down