From 42ffba11ad4597289b5ae609900a74a153fbd067 Mon Sep 17 00:00:00 2001
From: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Date: Wed, 11 Sep 2024 23:13:14 -0700
Subject: [PATCH] [Misc] Use RoPE cache for MRoPE (#8396)

---
 vllm/model_executor/layers/rotary_embedding.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/vllm/model_executor/layers/rotary_embedding.py b/vllm/model_executor/layers/rotary_embedding.py
index 7fa6c5e7fcde4..d4e9ed87ed54f 100644
--- a/vllm/model_executor/layers/rotary_embedding.py
+++ b/vllm/model_executor/layers/rotary_embedding.py
@@ -990,7 +990,7 @@ def get_rope(
                 base, is_neox_style, dtype, short_factor, long_factor,
                 **extra_kwargs)
         elif scaling_type == "mrope":
-            return MRotaryEmbedding(
+            rotary_emb = MRotaryEmbedding(
                 head_size,
                 rotary_dim,
                 max_position,