Skip to content

[Installation]: vllm-ascend单卡部署多模型报错 #4404

@yangqianjing

Description

@yangqianjing

Your current environment

(EngineCore_DP0 pid=634) INFO 11-22 10:01:28 [worker_v1.py:256] Available memory: 0, total memory: 65452113920
一张910B上部署Qwen3-vl-8b和qwen2.5-vl-3b的时候,8b的模型直接就可以启动,启动之后显存还有34g,但是我以同样的命令去启动3b的时候会报错显存为0

How you are installing vllm and vllm-ascend

pip install -vvv vllm vllm-ascend

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions