diff --git a/python/sglang/srt/layers/attention/nsa/nsa_indexer.py b/python/sglang/srt/layers/attention/nsa/nsa_indexer.py
index 7518c909a..181b29c8b 100644
--- a/python/sglang/srt/layers/attention/nsa/nsa_indexer.py
+++ b/python/sglang/srt/layers/attention/nsa/nsa_indexer.py
@@ -1003,7 +1003,7 @@ class Indexer(CustomOp):
         )  # [bs, 64, 64 + 64]
 
         q_pe = q_pe.view(bs, self.n_heads, 1, self.rope_head_dim)
-        q_pe = torch_npu.npu_interleave_rope(q_pe, cos, sin).view(
+        q_pe = torch_npu.npu_rotary_mul(q_pe, cos, sin).view(
             bs, self.n_heads, self.rope_head_dim
         )  # [bs, n, d]
         q = torch.cat([q_pe, q_nope], dim=-1)
@@ -1017,7 +1017,7 @@ class Indexer(CustomOp):
         )  # [bs, 64 + 64]
 
         k_pe = k_pe.view(-1, 1, 1, self.rope_head_dim)
-        k_pe = torch_npu.npu_interleave_rope(k_pe, cos, sin).view(
+        k_pe = torch_npu.npu_rotary_mul(k_pe, cos, sin).view(
             bs, 1, self.rope_head_dim
         )  # [bs, 1, d]
         k = torch.cat([k_pe, k_nope.unsqueeze(1)], dim=-1)  # [bs, 1, 128]
@@ -1087,7 +1087,7 @@ class Indexer(CustomOp):
         past_key_states = forward_batch.token_to_kv_pool.get_index_k_buffer(layer_id)
 
         x = x.view(-1, self.hidden_size)
-        weights = self.weights_proj(x.float())[0]
+        weights = self.weights_proj(x.float())[0].to(torch.bfloat16)
         block_table = (
             block_table[: actual_seq_lengths_q.size()[0]] if is_prefill else block_table
         )
diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py
index f849b7f48..31348b5be 100644
--- a/python/sglang/srt/models/deepseek_v2.py
+++ b/python/sglang/srt/models/deepseek_v2.py
@@ -3653,9 +3653,10 @@ class DeepseekV2ForCausalLM(nn.Module):
                 self_attn.w_kc = bind_or_assign(
                     self_attn.w_kc, w_kc.transpose(1, 2).contiguous().transpose(1, 2)
                 )
-                self_attn.w_vc = bind_or_assign(
-                    self_attn.w_vc, w_vc.contiguous().transpose(1, 2)
-                )
+                w_vc = w_vc.contiguous().transpose(1, 2)
+                if _is_npu:
+                    w_vc = w_vc.contiguous()
+                self_attn.w_vc = bind_or_assign(self_attn.w_vc, w_vc)
                 if (
                     hasattr(self_attn.kv_b_proj, "weight_scale")
                     and self_attn.w_scale is None
diff --git a/test/srt/ascend/test_ascend_deepep.py b/test/srt/ascend/test_ascend_deepep.py
index d6a94890a..f8cf760d5 100644
--- a/test/srt/ascend/test_ascend_deepep.py
+++ b/test/srt/ascend/test_ascend_deepep.py
@@ -47,6 +47,8 @@ class TestAscendDeepEP(CustomTestCase):
             1,
             "--ep-size",
             16,
+            "--max-running-requests",
+            24,
             "--moe-a2a-backend",
             "deepep",
             "--deepep-mode",