Skip to content

fixed: Modified the topkgating function and modified the test_moe file for testing #7163

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

Open
wants to merge 4 commits into
base: master
Choose a base branch
from
Open

fixed: Modified the topkgating function and modified the test_moe file for testing #7163

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
f4a43cb
fixed: Modified the topkgating function and modified the test_moe fil…
xiongjyu Mar 21, 2025
6123c74
style: fix formatting issue in test_moe.py
xiongjyu Mar 22, 2025
07a8649
Merge branch 'master' into master
loadams Mar 27, 2025
c030f49
Merge branch 'master' into master
xiongjyu Mar 28, 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
1 change: 1 addition & 0 deletions deepspeed/moe/sharded_moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -429,6 +429,7 @@ def topkgating(
tp = 1 if groups.mpu is None else bwc_tensor_model_parallel_world_size(mpu=groups.mpu)
new_capacity = torch.ceil(new_capacity / tp).mul(tp).to(new_capacity.dtype)
capacity = new_capacity
locations = torch.cumsum(mask, dim=0) - 1

# normalize gates
gates_masked = gates * mask
Expand Down
9 changes: 9 additions & 0 deletions tests/unit/moe/test_moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -241,6 +241,15 @@ def check_equal(logits, cap, sparse_truth, res):
position_dispatch_res = topkgating(logits2, 3, 1, min_capacity=1, drop_policy='position')[2]
check_equal(logits2, 2, position_sec_sparse, position_dispatch_res)

#s=4 e=4 topk=2 drop_tokens=False
logits3 = torch.tensor([[0.95, 0.85, 0.90, 0.80], [0.70, 0.65, 0.75, 0.60], [0.50, 0.55, 0.45, 0.40],
[0.35, 0.30, 0.25, 0.20]])
logits3 *= dist.get_rank() + 1
dispatch_res = topkgating(logits3, 2, 1, min_capacity=1, drop_tokens=False)[2]
sec_sparse = torch.tensor([[0, 0, 0], [0, 2, 0], [1, 0, 1], [1, 2, 1], [2, 0, 2], [2, 1, 0], [3, 0, 3],
[3, 1, 1]])
check_equal(logits3, 4, sec_sparse, dispatch_res)


class TestExpertWeightGradWithZero(DistributedTest):
world_size = 2
Expand Down