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 17, 04:15 UTC
Fix lint (#3388)
Antoni Baum committed
Mar 13, 2024 at 13:56 UTC
c33afd89f56ba5c260275fdd6723c59642f82f22
1 file changed
+2
-2
vllm/model_executor/layers/rotary_embedding.py
+2
-2
@@ -143,8 +143,8 @@ class RotaryEmbedding(nn.Module):
143
offsets: Optional[torch.Tensor] = None,
144
) -> Tuple[torch.Tensor, torch.Tensor]:
145
self.cos_sin_cache = self.cos_sin_cache.to(positions.get_device())
146
- # ops.rotary_embedding()/batched_rotary_embedding() are in-place operations that
147
- # update the query and key tensors.
146
+ # ops.rotary_embedding()/batched_rotary_embedding()
147
+ # are in-place operations that update the query and key tensors.
148
if offsets is not None:
149
ops.batched_rotary_embedding(positions, query, key, self.head_size,
150
self.cos_sin_cache,