diff --git a/vllm_omni/model_executor/models/fish_speech/fish_speech_fast_ar.py b/vllm_omni/model_executor/models/fish_speech/fish_speech_fast_ar.py
index cc38c200a..df3017152 100644
--- a/vllm_omni/model_executor/models/fish_speech/fish_speech_fast_ar.py
+++ b/vllm_omni/model_executor/models/fish_speech/fish_speech_fast_ar.py
@@ -106,7 +106,7 @@ class _FastARAttention(nn.Module):
             q = self.q_norm(q.view(-1, self.num_heads, self.head_dim)).view(q.shape)
             k = self.k_norm(k.view(-1, self.num_kv_heads, self.head_dim)).view(k.shape)
 
-        q, k = self.rotary_emb(position_ids, q, k)
+        q, k = self.rotary_emb(position_ids.reshape(-1), q, k)
 
         q = q.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2)
         k = k.view(bsz, seq_len, self.num_kv_heads, self.head_dim).transpose(1, 2)