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 15, 23:30 UTC
[Bugfix] Fix Qwen2_5_VLForConditionalGeneration packed_modules_mapping (#12905)
Jee Jee Li committed
Feb 8, 2025 at 13:02 UTC
306923da82535593bc508cc4e039bdec55159e9f
1 file changed
+5
-2
vllm/model_executor/models/qwen2_5_vl.py
+5
-2
@@ -760,9 +760,12 @@ class Qwen2_5_VLForConditionalGeneration(nn.Module, SupportsMultiModal,
760
"q_proj",
761
"k_proj",
762
"v_proj",
763
- ]
763
+ ],
764
+ "gate_up_proj": [
765
+ "gate_proj",
766
+ "up_proj",
767
+ ],
768
}
765
-
769
# LoRA specific attributes, TODO: double check
770
supported_lora_modules = [
771
"qkv_proj",