diff --git a/tests/ut/ops/test_rotary_embedding.py b/tests/ut/ops/test_rotary_embedding.py
index 3b388e09f29..c331015467b 100644
--- a/tests/ut/ops/test_rotary_embedding.py
+++ b/tests/ut/ops/test_rotary_embedding.py
@@ -219,11 +219,8 @@ def test_native_rope_deepseek_forward_cache_handling(
 
         mock_rope_forward_oot.return_value = (query, key)
 
-        q_pe, k_pe = native_rope_deepseek_forward(module,
-                                                  positions,
-                                                  query,
-                                                  key,
-                                                  max_seq_len=2048)
+        q_pe, k_pe = native_rope_deepseek_forward(module, positions, query,
+                                                  key)
 
         assert q_pe.shape == query.shape
         assert k_pe.shape == key.shape
diff --git a/vllm_ascend/attention/mla_v1.py b/vllm_ascend/attention/mla_v1.py
index a8f8ae82332..889ce4093b3 100644
--- a/vllm_ascend/attention/mla_v1.py
+++ b/vllm_ascend/attention/mla_v1.py
@@ -1186,9 +1186,7 @@ def forward(
             else:
                 decode_q_pe[...], decode_k_pe[...] = self.rotary_emb(
                     attn_metadata.decode.input_positions,
-                    decode_q_pe.contiguous(),
-                    decode_k_pe,
-                    max_seq_len=attn_metadata.decode.max_seq_lens)
+                    decode_q_pe.contiguous(), decode_k_pe)
         if has_prefill:
             assert attn_metadata.prefill is not None
             prefill_q = self.q_proj(prefill_hs_or_q_c)[0]\
@@ -1213,9 +1211,7 @@ def forward(
             else:
                 prefill_q_pe[...], prefill_k_pe[...] = self.rotary_emb(
                     attn_metadata.prefill.input_positions,
-                    prefill_q_pe.contiguous(),
-                    prefill_k_pe,
-                    max_seq_len=attn_metadata.prefill.max_seq_lens)
+                    prefill_q_pe.contiguous(), prefill_k_pe)
 
         assert len(
             kv_cache
diff --git a/vllm_ascend/ops/rotary_embedding.py b/vllm_ascend/ops/rotary_embedding.py
index 3dd91ea63f2..831361ca296 100644
--- a/vllm_ascend/ops/rotary_embedding.py
+++ b/vllm_ascend/ops/rotary_embedding.py
@@ -90,10 +90,7 @@ def native_rope_deepseek_forward(self,
                                  positions: torch.Tensor,
                                  query: torch.Tensor,
                                  key: torch.Tensor,
-                                 offsets: Optional[torch.Tensor] = None,
-                                 max_seq_len: Optional[int] = None):
-    if max_seq_len is not None and max_seq_len > self.max_seq_len:
-        _set_cos_sin_cache(self, max_seq_len, query.device, query.dtype)
+                                 offsets: Optional[torch.Tensor] = None):
     if len(key.shape) == 2:
         key = key[:, None, :]
     # Note: we implement the non neox_style method with shuffle the last dim and neox style
@@ -208,8 +205,7 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids, unsqueeze_dim=1):
     return q_embed, k_embed
 
 
-def _set_cos_sin_cache(self, seq_len, device, dtype):
-    self.max_seq_len_cached = seq_len
+def _set_cos_sin_cache(self, max_seq_len, device, dtype):
     dim = self.rotary_dim
 
     freq_extra = 1.0 / (self.base**(
@@ -229,9 +225,7 @@ def _set_cos_sin_cache(self, seq_len, device, dtype):
     inv_freq = freq_inter * (1 - inv_freq_mask) + freq_extra * inv_freq_mask
     self.register_buffer("inv_freq", inv_freq, persistent=False)
 
-    t = torch.arange(seq_len * self.scaling_factor,
-                     device=device,
-                     dtype=torch.float32)
+    t = torch.arange(max_seq_len, device=device, dtype=torch.float32)
 
     freqs = torch.outer(t, inv_freq)
     cos_cached = torch.cat([freqs, freqs], dim=-1).cos() * self.mscale
@@ -276,11 +270,10 @@ def deepseek_rope_init_func(
     super(DeepseekScalingRotaryEmbedding,
           self).__init__(head_size, rotary_dim, max_position_embeddings, base,
                          is_neox_style, dtype)
-    self.max_seq_len = max_position_embeddings
-    _set_cos_sin_cache(self,
-                       max_position_embeddings,
-                       dtype=dtype,
-                       device="npu")
+
+    # NOTE: For ascend friendly computing, reorder sin and cos cache
+    self.max_seq_len = max_position_embeddings * scaling_factor
+    _set_cos_sin_cache(self, self.max_seq_len, dtype=dtype, device="npu")
 
 
 RotaryEmbedding.forward_oot = rope_forward_oot
diff --git a/vllm_ascend/worker/model_runner_v1.py b/vllm_ascend/worker/model_runner_v1.py
index ba1657c611b..7c04b73d07e 100644
--- a/vllm_ascend/worker/model_runner_v1.py
+++ b/vllm_ascend/worker/model_runner_v1.py
@@ -345,6 +345,8 @@ def __init__(self, vllm_config: VllmConfig, device: torch.device):
         self.torchair_graph_enabled = ascend_config.torchair_graph_config.enabled
         self.use_cached_npu_graph = ascend_config.torchair_graph_config.use_cached_graph
         self.torchair_graph_batch_sizes = ascend_config.torchair_graph_config.graph_batch_sizes
+        self.use_ring_mla = ascend_config.chunked_prefill_for_mla
+
         if ascend_config.torchair_graph_config.graph_batch_sizes_init:
             self.init_torchair_graph_batch_sizes()
 
@@ -1137,11 +1139,14 @@ def _process_reqs(
         else:
             attn_state = AscendAttentionState.PrefillCacheHit
 
-        self.attn_mask = self._make_attention_mask(
-            seq_lens=seq_lens,
-            query_lens=num_scheduled_tokens,
-            position=positions,
-            attn_state=attn_state)
+        # NOTE: when use ring_mla, attn_mask don't need to generate here.
+        if not self.use_ring_mla or attn_state == AscendAttentionState.PrefillNoCache:
+            attn_mask = self._make_attention_mask(
+                seq_lens=seq_lens,
+                query_lens=num_scheduled_tokens,
+                position=positions,
+                attn_state=attn_state)
+            self.attn_mask = attn_mask
         self.attn_state = attn_state  # type: ignore
 
         extra_builder_kwargs = {}