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 10, 04:00 UTC
[Misc] Remove redundant line (#32366)
Signed-off-by: wangli <wangli858794774@gmail.com>
Li Wang committed
Jan 15, 2026 at 12:29 UTC
d86fc23bdd8a67a0f466d73f7ea6c50871abe232
1 file changed
-1
vllm/attention/layer.py
-1
@@ -363,7 +363,6 @@ class Attention(nn.Module, AttentionLayerBase):
363
output_shape = torch.Size(
364
(num_tokens, self.num_heads * self.head_size_v)
365
)
366
- output_shape = output_shape if output_shape is not None else query.shape
366
output = torch.empty(output_shape, dtype=output_dtype, device=query.device)
367
hidden_size = output_shape[-1]
368
# Reshape the query, key, and value tensors.