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 5, 13:30 UTC
[Bugfix][LoRA] Fix missing in_proj_z in Qwen3_5ForConditionalGenerati… (#38927)
Signed-off-by: elenalil-aws <elenalil@amazon.com>
elenalil-aws committed
Apr 3, 2026 at 16:30 UTC
81994e1d0ea6e0486cfbdad02e00942614c55264
1 file changed
+1
vllm/model_executor/models/qwen3_5.py
+1
@@ -620,6 +620,7 @@ class Qwen3_5ForConditionalGeneration(Qwen3VLForConditionalGeneration, IsHybrid)
620
self.packed_modules_mapping = {k: list(v) for k, v in base.items()}
621
self.packed_modules_mapping.pop("in_proj_qkvz", None)
622
self.packed_modules_mapping["in_proj_qkv"] = ["in_proj_qkv"]
623
+ self.packed_modules_mapping["in_proj_z"] = ["in_proj_z"]
624
625
def embed_input_ids(
626
self,