[Bug]: vllm server 部署base和lora模型后，请求lora模型失败 about vllm HOT 3 OPEN

Which version of vLLM are you using? Additionally, could you provide more detailed information about the error you're encountering?

from vllm.

Which version of vLLM are you using? Additionally, could you provide more detailed information about the error you're encountering?
vllm=0.5.4
具体信息：我用llama-factory这个项目，lora训练了 qwen2-7b-instruct，借鉴与 longlora，训练参数里还有norm和embed。之后用vllm部署 模型 和 lora（包含norm 和 embed）
vllm serve /root/autodl-fs/llm_models/qwen/Qwen2-7B-Instruct --enable-lora --lora-modules bi-lora=/root/autodl-fs/saves/Qwen2-7B-Instruct/lora/sft/checkpoint-1500/ --port 6006
上面的命令里 /root/autodl-fs/saves/Qwen2-7B-Instruct/lora/sft/checkpoint-1500/ 是lora参数（包含norm和embed）所在的目录。
访问的时候，下面的请求体会报错：RuntimeError: Loading lora /root/autodl-fs/saves/Qwen2-7B-Instruct/lora/sft/checkpoint-1500/ failed 。但访问 chat模型是没问题的（model参数改成 /root/autodl-fs/llm_models/qwen/Qwen2-7B-Instruct）
{
"model": "bi-lora",
"prompt": "你是谁",
"temperature": 0.7,
"top_p": 0.8,
"repetition_penalty": 1.05,
"max_tokens": 512
}

from vllm.

The cause of this error is that it caught an execution error from the previous code, so there should be error messages preceding this one. You should check the log information again.

from vllm.