[Misc] Use RoPE cache for MRoPE (#8396)

42ffba11 · Woosuk Kwon · GitHub · 295c4730 · 42ffba11
Unverified Commit 42ffba11 authored Sep 11, 2024 by Woosuk Kwon Committed by GitHub Sep 11, 2024
Hide whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

vllm/model_executor/layers/rotary_embedding.py vllm/model_executor/layers/rotary_embedding.py +1 -1

No files found.
--- 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,