From c1d1412333f220812c679855336ae9144efcb3c5 Mon Sep 17 00:00:00 2001 From: Baizhou Zhang Date: Mon, 27 Apr 2026 20:37:56 -0700 Subject: [PATCH] [Flashinfer] Integrate flashinfer router gemm for sm103 (#23285) --- 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 b226dc3b9..475fbda53 100644 --- a/python/sglang/srt/models/deepseek_v2.py +++ b/python/sglang/srt/models/deepseek_v2.py @@ -335,7 +335,7 @@ 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 in [100, 103] and self.weight.shape[0] == 256: # router gemm output float32 logits = torch.empty( hidden_states.shape[0],