Skip to content

[Bugfix][Model] Fix Ling3 SwiGLU clamping #65088

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
vllm-bot merged 3 commits into from
Nov 5, 2026
Merged

[Bugfix][Model] Fix Ling3 SwiGLU clamping #65088

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c417f2d
[Bugfix][Model] Fix Ling SwiGLU clamping
zexplorerhj Nov 2, 2026
a84c95b
Merge branch 'main' into fix/ling3-post-silu-limit
zexplorerhj Nov 2, 2026
3398af7
Merge branch 'main' into fix/ling3-post-silu-limit
zexplorerhj Nov 2, 2026
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
5 changes: 3 additions & 2 deletions vllm/model_executor/layers/fused_moe/activation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -272,7 +272,7 @@ def _apply_moe_activation_masked(
"SWIGLUOAI_UNINTERLEAVE requires clamp_limit"
)
elif activation == MoEActivation.SWIGLUSTEP:
masked_clamp_limit = 7.0
masked_clamp_limit = 7.0 if config.clamp_limit is None else config.clamp_limit

torch.ops._C.masked_moe_activation(
output,
Expand Down Expand Up @@ -390,7 +390,8 @@ def apply_moe_activation(
elif activation == MoEActivation.SWIGLUSTEP:
from vllm.model_executor.layers.activation import swiglustep_and_mul_triton

swiglustep_and_mul_triton(output, input)
limit = 7.0 if config.clamp_limit is None else config.clamp_limit
swiglustep_and_mul_triton(output, input, limit=limit)

# Activations without gated multiplication
elif activation == MoEActivation.SILU_NO_MUL:
Expand Down
4 changes: 3 additions & 1 deletion vllm/model_executor/models/bailing_moe_v3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1203,7 +1203,9 @@ def __init__(
intermediate_size=config.moe_intermediate_size,
renormalize=True,
quant_config=quant_config,
activation="silu",
activation=(
"swiglustep" if self.expert_swiglu_limit is not None else "silu"
),
swiglu_limit=self.expert_swiglu_limit,
prefix=f"{prefix}.experts",
scoring_func="sigmoid",
Expand Down