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 14, 05:30 UTC
Remove Redundant Assignment in Qwen3_VisionPatchMerger (#25224)
Signed-off-by: Junhong <liujunhong11@huawei.com> Co-authored-by: Junhong <liujunhong11@huawei.com> Co-authored-by: Roger Wang <hey@rogerw.io>
LJH-LBJ committed
Sep 20, 2025 at 02:15 UTC
d90e212a3a586b8a6ca9a424868abd0e6ef6779a
1 file changed
+1
-3
vllm/model_executor/models/qwen3_vl.py
+1
-3
@@ -223,9 +223,7 @@ class Qwen3_VisionPatchMerger(nn.Module):
223
224
if norm_layer is None:
225
norm_layer = partial(nn.LayerNorm, eps=1e-6)
226
- self.use_postshuffle_norm = use_postshuffle_norm
227
- self.norm = norm_layer(
228
- self.hidden_size if use_postshuffle_norm else context_dim)
226
+ self.norm = norm_layer(context_dim)
227
self.linear_fc1 = ColumnParallelLinear(self.hidden_size,
228
self.hidden_size,
229
bias=True,