moe support clamp#395
Merged
Merged
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
Note
Copilot was unable to run its full agentic suite in this review.
Adds an optional GEMM1 (gate/up) clamping feature to fused MoE, intended to limit activation magnitude before the activation/multiply step.
Changes:
- Extend
ref_fused_moe,FusedMoeinit, andxpu_fused_moepublic API to acceptgemm1_clamp_limit. - Apply in-place clamping to gate/up outputs before activation in both reference and kernel paths.
- Thread the new parameter through
_apply_refand kernel execution paths.
Comments suppressed due to low confidence (1)
vllm_xpu_kernels/fused_moe_interface.py:542
- The public API adds
gemm1_clamp_limitbut the function docstring doesn’t describe it (expected type/units, what exactly is clamped, and that gate is clamped only on max while up is clamped symmetrically). Please documentgemm1_clamp_limitbehavior here (and anywhere else this is a public/configurable parameter) so callers understand the numerical impact and intended use.
def xpu_fused_moe(hidden_states,
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+562
to
566
| is_block_fp8=False, | ||
| gemm1_clamp_limit=None): | ||
| ''' | ||
| hidden_states: [num_rows, hidden_size] | ||
| w13: [num_experts, 2*inter_size, hidden_size] |
Comment on lines
+506
to
+507
| # Apply swiglu_limit clamping before activation | ||
| if self.gemm1_clamp_limit is not None and self.gemm1_clamp_limit > 0: |
Comment on lines
+506
to
+511
| # Apply swiglu_limit clamping before activation | ||
| if self.gemm1_clamp_limit is not None and self.gemm1_clamp_limit > 0: | ||
| gate = gemm1_output[:, :self.inter_size] | ||
| up = gemm1_output[:, self.inter_size:] | ||
| gate.clamp_(max=self.gemm1_clamp_limit) | ||
| up.clamp_(min=-self.gemm1_clamp_limit, max=self.gemm1_clamp_limit) |
xinyu-intel
reviewed
Jun 5, 2026
xinyu-intel
left a comment
xinyu-intel
left a comment
Collaborator
There was a problem hiding this comment.
please add a test case.
xinyu-intel
added
the
API-compatible
wuxun-zhang
approved these changes
Jun 5, 2026
wuxun-zhang
left a comment
wuxun-zhang
left a comment
Contributor
There was a problem hiding this comment.
LGTM. It aligns with DeepSeek v4 model config and what NV/AMD did: https://github.com/vllm-project/vllm/blob/main/vllm/models/deepseek_v4/nvidia/model.py#L466.
Maybe later we can fuse this clamp with activation kernel.
majian4work
force-pushed
the
fix-moe
branch
2 times, most recently
from
01a6c8a to
09160ea
Compare
xinyu-intel
approved these changes
Jun 5, 2026
jikunshang
approved these changes
Jun 8, 2026
jikunshang
left a comment
jikunshang
left a comment
Member
There was a problem hiding this comment.
LGTM.
cc @mayuyuace PTAL, thanks.
mayuyuace
approved these changes
Jun 8, 2026
majian4work
force-pushed
the
fix-moe
branch
2 times, most recently
from
e3b3082 to
4125b47
Compare
Signed-off-by: Ma Jian <jian1.ma@intel.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
DSV4 needs a clamp before activation.