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, 10:15 UTC
[Bugfix] Fix typo in FusedMoE LoRA reshape comment (#31992)
Signed-off-by: Xin Yang <xyangx@amazon.com>
Xin Yang committed
Jan 8, 2026 at 18:46 UTC
8413868dab334b23cf8684671eccda3fdd96cc70
1 file changed
+1
-1
vllm/lora/model_manager.py
+1
-1
@@ -632,7 +632,7 @@ class LoRAModelManager:
632
num_experts, -1, down_proj_lora.lora_a.shape[-1]
633
)
634
635
- # (output_size,num_experts,rank)
635
+ # (output_size,rank,num_experts)
636
gate_up_proj_lora.lora_b = gate_up_proj_lora.lora_b.reshape(
637
gate_up_proj_lora.lora_b.shape[0], -1, num_experts
638
)