diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py
index 2b87d91d4..bb1efde29 100644
--- a/python/sglang/srt/models/deepseek_v2.py
+++ b/python/sglang/srt/models/deepseek_v2.py
@@ -1154,7 +1154,7 @@ class DeepseekV2AttentionMLA(nn.Module):
         _, q_pe = q.split([self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1)
         kv_a, _ = latent_cache.split([self.kv_lora_rank, self.qk_rope_head_dim], dim=-1)
         latent_cache = latent_cache.unsqueeze(1)
-        kv_a = self.kv_a_layernorm(kv_a.contiguous())
+        kv_a = self.kv_a_layernorm(kv_a)
         kv = self.kv_b_proj(kv_a)[0]
         kv = kv.view(-1, self.num_local_heads, self.qk_nope_head_dim + self.v_head_dim)
         k_nope = kv[..., : self.qk_nope_head_dim]
@@ -1693,7 +1693,7 @@ class DeepseekV2AttentionMLA(nn.Module):
         _, q_pe = q.split([self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1)
         kv_a, _ = latent_cache.split([self.kv_lora_rank, self.qk_rope_head_dim], dim=-1)
         latent_cache = latent_cache.unsqueeze(1)
-        kv_a = self.kv_a_layernorm(kv_a.contiguous())
+        kv_a = self.kv_a_layernorm(kv_a)
         kv = self.kv_b_proj(kv_a)[0]
         kv = kv.view(-1, self.num_local_heads, self.qk_nope_head_dim + self.v_head_dim)
         k_nope = kv[..., : self.qk_nope_head_dim]