Skip to content

fix legacy torch save when tensor_model_parallel_size > expert_model_parallel_size * expert_tensor_parallel_size #4678

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
dimapihtar merged 6 commits into from
May 13, 2026
Merged

fix legacy torch save when tensor_model_parallel_size > expert_model_parallel_size * expert_tensor_parallel_size #4678

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
445e107
fix docs
dimapihtar May 5, 2026
106a0be
Merge branch 'NVIDIA:main' into main
dimapihtar May 6, 2026
422590f
Merge branch 'NVIDIA:main' into main
dimapihtar May 6, 2026
818fd81
Merge branch 'NVIDIA:main' into main
dimapihtar May 7, 2026
0d4e12e
fix moe torch save when TP > EP * ETP
dimapihtar May 7, 2026
0f64f7b
Merge branch 'main' into fix_moe_torch_ckpt
dimapihtar May 7, 2026
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
5 changes: 5 additions & 0 deletions megatron/training/checkpointing.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -596,7 +596,12 @@ def save_checkpoint(iteration, model, optimizer, opt_param_scheduler, num_floati
rank = torch.distributed.get_rank() if torch.distributed.is_initialized() else 0

# Collect args, model, RNG.
# For LEGACY checkpoints, every unique (tp_rank, ep_rank) shard must be written by
# exactly one rank. Neither dp_rank==0 nor edp_rank==0 alone covers all shards when
# the dense and expert parallelism layouts disagree (e.g. TP > EP*ETP); the union
# does, with at most one rank per (tp_rank, ep_rank) inside any DP group.
if not torch.distributed.is_initialized() \
or mpu.get_data_parallel_rank() == 0 \
or mpu.get_expert_data_parallel_rank() == 0 \
or ckpt_type != CheckpointType.LEGACY:
if ckpt_type != CheckpointType.LEGACY:
Expand Down
Loading