Skip to content

Change mosaic_group_id in case of GPUs/CPUs #3041

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
copybara-service merged 1 commit into from
Feb 2, 2026
Merged

Change mosaic_group_id in case of GPUs/CPUs #3041

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: 5 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 @@ -950,9 +950,13 @@ def gmm(inputs, kernel, tiling, group_sizes, expert_assignments, weight_gather_a
# Use full contraction for QWIX quantization to allow quantization
# fusion (max reduce over contracting dimension).
tiling = (tiling[0], k, tiling[2])

is_tpu = (self.mesh.devices.flat[0] == "tpu")
# TPU needs random mosaic_fusion_group; GPU/CPU needs deterministic ID for autotuner sync
mosaic_group_id = f"{random.randint(0, 1000000000)}" if is_tpu else '0'
with set_xla_metadata(
ragged_dot_tiling=",".join([str(t) for t in tiling]),
mosaic_fusion_group=f"{random.randint(0, 1000000000)}",
mosaic_fusion_group=mosaic_group_id,
):
output = jax.lax.ragged_dot(
lhs=inputs,
Expand Down
Loading