Gemma4 is not working(serverless) via vLLM v2.14.0.

[south sundial]

Workers just can't start.

[dense salmon]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[spark elm]

Yup vvlm worker has not yet been updated to latest upstream vllm

[south sundial]

can you help me with Network venue ? I can't setup it, because I don't know on which data centers is available for my GPU

[south sundial]

I configured it this way, but it still doesn’t behave as expected. My goal is to send a single request and have the service quickly start inference from the disk image without a long cold start.

[hexed sparrow]

what are your env variables

can i see your env variables, all of them

dont screenhot tokens if you have any ( blur / cover them)

usually bigger models still do take time to load so its normal

[hexed sparrow]

seems normal to me, i think its just the loading time

[south sundial]

so it's should be the best setup, as I can do? with cold start?

where can I see Network volumes ? I mean that how much space there is free or occupied