Skip to content

set pg names #1910

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

Closed
wants to merge 2 commits into from
Closed

set pg names #1910

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
22a1a9a
fix setting ft state dicts when ft checkpointing is disabled
tushar00jain Oct 29, 2025
7e6afe5
set pg names
tushar00jain Oct 29, 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
Prev Previous commit
set pg names
  • Loading branch information
@tushar00jain
tushar00jain committed Oct 29, 2025
commit 7e6afe5c2eb4dbd6d02f16f19f3170390c2b912c
3 changes: 2 additions & 1 deletion torchtitan/distributed/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -232,7 +232,7 @@ def maybe_enable_amp(


def init_distributed(
comm_config: CommConfig, enable_cpu_backend: bool = False, base_folder: str = ""
comm_config: CommConfig, enable_cpu_backend: bool = False, base_folder: str = "", ranks: list[int] = []
):
def _warn_overwrite_env(env, val):
if env in os.environ:
Expand Down Expand Up @@ -276,6 +276,7 @@ def _get_distributed_backend(enable_cpu_backend):
torch.distributed.init_process_group(
backend=_get_distributed_backend(enable_cpu_backend),
timeout=timedelta(seconds=comm_config.init_timeout_seconds),
_ranks=ranks,
)


Expand Down
10 changes: 10 additions & 0 deletions torchtitan/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -83,11 +83,21 @@ def __init__(self, job_config: JobConfig):
# Device has to be set before creating TorchFT manager.
device_module.set_device(self.device)

ranks = []
ft_config = job_config.fault_tolerance
if ft_config.enable:
group_size = ft_config.group_size
replica_id = ft_config.replica_id
first_rank = replica_id * group_size
last_rank = first_rank + group_size - 1
ranks = list(range(first_rank, last_rank + 1))

# init distributed and build meshes
dist_utils.init_distributed(
job_config.comm,
enable_cpu_backend=job_config.training.enable_cpu_offload,
base_folder=job_config.job.dump_folder,
ranks=ranks,
)

job_config.maybe_log()
Expand Down
Loading