From c5fa364b80b365fc916279d13c6d9384a65b9b3c Mon Sep 17 00:00:00 2001 From: Baizhou Zhang Date: Sun, 5 Apr 2026 09:33:14 -0700 Subject: [PATCH] [Hotfix] Fix router gemm on sm103 (#22134) --- python/sglang/srt/models/deepseek_v2.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py index 87d751465..24b46c0bc 100644 --- a/python/sglang/srt/models/deepseek_v2.py +++ b/python/sglang/srt/models/deepseek_v2.py @@ -326,7 +326,7 @@ class MoEGate(nn.Module): and (self.weight.shape[0] == 256 or self.weight.shape[0] == 384) and _device_sm >= 90 ): - if _device_sm >= 100 and self.weight.shape[0] == 256: + if _device_sm == 100 and self.weight.shape[0] == 256: # router gemm output float32 logits = torch.empty( hidden_states.shape[0],