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 2, 01:00 UTC
[LoRA] Fix dedup for post-replacement module aliases (#44413)
Signed-off-by: bk-201 <joy25810@foxmail.com>
linitra24 committed
Jun 4, 2026 at 02:23 UTC
271328e256bae5d5728eea32eccf44c91ad5f147
1 file changed
+1
vllm/lora/model_manager.py
+1
@@ -451,6 +451,7 @@ class LoRAModelManager:
451
)
452
if isinstance(new_module, BaseLayerWithLoRA):
453
wrapped_by_id[id(module)] = new_module
454
+ wrapped_by_id[id(new_module)] = new_module
455
456
# (yard1): TODO make this more robust
457
if "lm_head" in module_name: