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 7, 22:45 UTC
Fix pipeline parallel with embed scaling in the Transformers modelling backend (#35094)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Feb 23, 2026 at 13:04 UTC
103e614b1487fd58477bbe7354a3cb2e9162e388
1 file changed
+1
vllm/model_executor/models/transformers/base.py
+1
@@ -191,6 +191,7 @@ class Base(
191
self.attention_instances = self.create_attention_instances()
192
193
# Input embeddings
194
+ self.embed_scale = None
195
input_embeddings = self.model.get_input_embeddings()
196
if not isinstance(input_embeddings, PPMissingLayer):
197
# Some models scale embeddings inside the input embedding layer