Skip to content

Update tokamax group sizes for pipeline #3397

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
NuojCheng wants to merge 1 commit into
base: main
Choose a base branch
from
+8 −4
Open

Update tokamax group sizes for pipeline #3397

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
12 changes: 8 additions & 4 deletions src/maxtext/layers/moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -896,10 +896,14 @@ def sparse_matmul(
def gmm(
inputs, kernel, tiling, group_sizes, expert_assignments, weight_gather_axes, input_buffer_count, combine_scopes
):
tokamax_group_sizes = tokamax.RaggedDotGroupSizes(
group_sizes,
max_utils.generate_representative_group_sizes(inputs.shape[0], kernel.shape[0]),
)
# 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:
tokamax_group_sizes = group_sizes
else:
tokamax_group_sizes = tokamax.RaggedDotGroupSizes(
group_sizes,
max_utils.generate_representative_group_sizes(inputs.shape[0], kernel.shape[0]),
)
pad_length = self.config.wi_tile_fwd_batch_seq
hs_shape = inputs.shape
# pad length is the 1st dimension of tiling size in gmm call
Expand Down
Loading