@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 42ffba11ad
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 16, 17:00 UTC

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

Woosuk Kwon committed Sep 11, 2024 at 23:13 UTC 42ffba11ad4597289b5ae609900a74a153fbd067
1 file changed +1 -1
vllm/model_executor/layers/rotary_embedding.py
+1 -1
@@ -990,7 +990,7 @@ def get_rope(
990 base, is_neox_style, dtype, short_factor, long_factor,
991 **extra_kwargs)
992 elif scaling_type == "mrope":
993 - return MRotaryEmbedding(
993 + rotary_emb = MRotaryEmbedding(
994 head_size,
995 rotary_dim,
996 max_position,