Skip to content

[mxfp8 moe training] update benchmarks to force load balancing #3193

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
merged 1 commit into from
Oct 17, 2025
Merged

[mxfp8 moe training] update benchmarks to force load balancing #3193

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
6 changes: 4 additions & 2 deletions benchmarks/prototype/moe_training/bench_moe_layer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,10 +23,10 @@

# this benchmark requires torchtitan
try:
from torchtitan.distributed.expert_parallel import (
from torchtitan.models.moe import MoE, MoEArgs
from torchtitan.models.moe.utils import (
set_token_group_alignment_size_m,
)
from torchtitan.models.moe import MoE, MoEArgs
except ImportError:
logging.warning(
"please pip install torchtitan to run this benchmark: https://github.com/pytorch/torchtitan"
Expand Down Expand Up @@ -77,6 +77,8 @@ def bench_moe_training_fsdp(args: argparse.Namespace):
target_fqns = ["experts"]
model_args = MoEArgs(
num_experts=local_num_experts,
num_shared_experts=1,
_debug_force_load_balance=True,
)
init_std = 0.02
device = torch.device("cuda")
Expand Down
6 changes: 4 additions & 2 deletions benchmarks/prototype/moe_training/benchmark_moe_layer_fsdp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,10 +32,10 @@

# this benchmark requires torchtitan
try:
from torchtitan.distributed.expert_parallel import (
from torchtitan.models.moe import MoE, MoEArgs
from torchtitan.models.moe.utils import (
set_token_group_alignment_size_m,
)
from torchtitan.models.moe import MoE, MoEArgs
except ImportError:
pytest.skip(
"torchtitan not installed, skipping MoE tests.", allow_module_level=True
Expand Down Expand Up @@ -71,6 +71,8 @@ def bench_moe_training_fsdp(recipe_name: str, enable_profile: bool, use_compile:
target_fqns = ["experts"]
model_args = MoEArgs(
num_experts=16,
num_shared_experts=1,
_debug_force_load_balance=True,
)
init_std = 0.02
device = torch.device("cuda")
Expand Down
Loading