diff --git a/vllm/model_executor/models/qwen3_5.py b/vllm/model_executor/models/qwen3_5.py index 3fb3b7a782..e8cc041ce4 100644 --- a/vllm/model_executor/models/qwen3_5.py +++ b/vllm/model_executor/models/qwen3_5.py @@ -309,6 +309,11 @@ class Qwen3_5ForCausalLMBase( "in_proj_qkvz": ["in_proj_qkv", "in_proj_z"], "in_proj_ba": ["in_proj_b", "in_proj_a"], } + # Maps PEFT embed/lm_head LoRA targets onto vLLM embedding wrappers. + embedding_modules = { + "embed_tokens": "input_embeddings", + "lm_head": "output_embeddings", + } # Some community text-only checkpoints keep the extraneous # `model.language_model.` prefix inherited from the VL training stack.