diff --git a/python/sglang/srt/models/deepseek_v4.py b/python/sglang/srt/models/deepseek_v4.py
index 4ab6d00e8..f9508ed95 100644
--- a/python/sglang/srt/models/deepseek_v4.py
+++ b/python/sglang/srt/models/deepseek_v4.py
@@ -623,6 +623,7 @@ class MQALayer(nn.Module):
                 o.reshape(T * G, D).contiguous(),
                 group_size=128,
             )
+            o_s = deep_gemm.ceil_to_ue8m0(o_s)
             output = torch.empty(T, G, R, device=o.device, dtype=torch.bfloat16)
             deep_gemm.fp8_einsum(
                 "bhr,hdr->bhd",