Skip to content

[FIX] Fix MoE GMM FP8 accumulator verification error and bump qwix requirement (Maxtext) #3507

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
sierraisland wants to merge 1 commit into
base: main
Choose a base branch
from
+8 −2
Open

[FIX] Fix MoE GMM FP8 accumulator verification error and bump qwix requirement (Maxtext) #3507

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
2 changes: 1 addition & 1 deletion requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ jaxtyping
fastapi[standard]<0.137.0
flax==0.12.8
torchax==0.0.13
qwix==0.1.2
qwix>=0.1.8
torchvision==0.25.0
pathwaysutils
parameterized
Expand Down
8 changes: 7 additions & 1 deletion tpu_inference/layers/common/fused_moe_gmm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,7 +108,12 @@ def gmm_wrapper(lhs,
group_offset,
fuse_act=None,
preferred_element_type=None):
gmm_res = gmm_v2(
try:

@wenxindongwork wenxindongwork Sep 2, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

With this change now all GMM accumulates in f32, i think this is will impact performance for our native model runs.

The real issue is that maxtext models do not properly support fp8 weights, so the fix really should exist on the maxtext side.

from qwix._src.interception import disable_interceptions
target_gmm = disable_interceptions(gmm_v2)
except Exception:
target_gmm = gmm_v2
gmm_res = target_gmm(
lhs=lhs,
rhs=rhs,
rhs_scale=rhs_scale,
Expand All @@ -118,6 +123,7 @@ def gmm_wrapper(lhs,
zero_initialize=False,
fuse_act=fuse_act,
preferred_element_type=preferred_element_type,
acc_dtype=jnp.float32,
)
return gmm_res

Expand Down
Loading