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: Apply embedding_multiplier to inputs_embeds (#34813)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
Gabe Goodhart committed
Feb 23, 2026 at 01:42 UTC
e631f8e78ef78fe6cf13903e27c827b45b25a0d0
1 file changed
+1
-1
vllm/model_executor/models/granitemoehybrid.py
+1
-1
@@ -378,7 +378,7 @@ class GraniteMoeHybridModel(nn.Module):
378
hidden_states = inputs_embeds
379
else:
380
hidden_states = self.embed_input_ids(input_ids)
381
- hidden_states = hidden_states * self.embedding_multiplier
381
+ hidden_states *= self.embedding_multiplier
382
residual = None
383
else:
384
if intermediate_tensors is None: