Skip to content

Fix backward compatibility issue with MFSDP --grad-reduce-in-bf16 #3799

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
shjwudp merged 2 commits into from
Mar 12, 2026
Merged

Fix backward compatibility issue with MFSDP --grad-reduce-in-bf16 #3799

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
15c2369
Fix forward compatibility issue with MFSDP --grad-reduce-in-bf16
shjwudp Mar 11, 2026
a89014b
Update megatron/training/arguments.py
shjwudp Mar 11, 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
3 changes: 3 additions & 0 deletions megatron/training/arguments.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -872,6 +872,9 @@ def validate_args(args, defaults={}):
args.megatron_fsdp_main_params_dtype = map_dtype(args.megatron_fsdp_main_params_dtype)
args.megatron_fsdp_main_grads_dtype = map_dtype(args.megatron_fsdp_main_grads_dtype)
args.megatron_fsdp_grad_comm_dtype = map_dtype(args.megatron_fsdp_grad_comm_dtype)
if args.grad_reduce_in_bf16:
assert args.megatron_fsdp_grad_comm_dtype == torch.bfloat16, \
"When --grad-reduce-in-bf16 is set, --megatron-fsdp-grad-comm-dtype must be bfloat16"

if args.fp8_param_gather:
assert args.use_distributed_optimizer or args.use_torch_fsdp2 or args.use_megatron_fsdp or not torch.is_grad_enabled(), \
Expand Down
Loading