Skip to content

Use non-tokamax group sizes for expert parallelism #3436

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

Draft
xuefgu wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Use non-tokamax group sizes for expert parallelism #3436

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
3 changes: 2 additions & 1 deletion src/maxtext/layers/moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -897,7 +897,8 @@ def gmm(
inputs, kernel, tiling, group_sizes, expert_assignments, weight_gather_axes, input_buffer_count, combine_scopes
):
# TODO (b/491979205) pipeline fsdp ag per repeat fails tokamax gmm
if self.config.using_pipeline_parallelism and self.config.pipeline_fsdp_ag_per_repeat:
# TODO (b/493621965) Expert parallelism fails tokamax gmm when vLLM uses MaxText models
if (self.config.using_pipeline_parallelism and self.config.pipeline_fsdp_ag_per_repeat) or self.config.rollout_expert_parallelism > 1:
tokamax_group_sizes = group_sizes
else:
tokamax_group_sizes = tokamax.RaggedDotGroupSizes(
Expand Down
Loading