[DSV4][XPU] Pass gemm1_clamp_limit to XpuFusedMoe (#44517)
Signed-off-by: Ma Jian <jian1.ma@intel.com>
Ma Jian committed
Jun 22, 2026 at 12:57 UTC
9037498c22891e55b594f567fb91d9b4efbf3e99
1 file changed
+2
vllm/model_executor/layers/fused_moe/experts/xpu_moe.py
+2
@@ -67,6 +67,7 @@ class XPUExperts(mk.FusedMoEExpertsModular):
67
self.is_mxfp4 = False
68
self.is_block_fp8 = False
69
self.is_mxfp8 = False
70
+ self.gemm1_clamp_limit = quant_config.gemm1_clamp_limit
71
self.fused_moe_impl: XpuFusedMoe | None = None
72
73
@property
@@ -176,6 +177,7 @@ class XPUExperts(mk.FusedMoEExpertsModular):
177
is_mxfp4=self.is_mxfp4,
178
is_mxfp8=self.is_mxfp8,
179
is_block_fp8=self.is_block_fp8,
180
+ gemm1_clamp_limit=self.gemm1_clamp_limit,
181
)
182
assert self.fused_moe_impl is not None
183
self.fused_moe_impl.apply(