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],