Skip to content

hub.json - Using the deploy form on Runpod vLLM forces the GPU_MEMORY_UTILIZATION field setting full numbers #278

@walhalla-software-dev

Description

@walhalla-software-dev

Using the UI on RunPod for deploying a serverless vLLM instance:

Default value is 0,95

Typing 0,92 e.g. sets automatically 1

I think there should be set a "step" or the issue is at RunPod itself interpreting the json

...
{ "key": "GPU_MEMORY_UTILIZATION", "input": { "name": "GPU Memory Utilization", "type": "number", "description": "Sets GPU VRAM utilization", "default": 0.9, "step": 0.01 } }

https://github.com/runpod-workers/worker-vllm/blob/main/.runpod/hub.json

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions