Runpod•2mo ago

vLLM v2.9.6 Serverless Stuck in Initializing and no workers

I have been trying to get the template up with multiple hf models

Latest I tried was https://huggingface.co/solidrust/L3-8B-Stheno-v3.2-AWQ

My endpoint is non stop in initializing but I see no workers booting up

solidrust/L3-8B-Stheno-v3.2-AWQ · Hugging Face

Tirth•11/7/25, 6:34 PM

Hi, can you DM your endpoint ID?

TTirth Hi, can you DM your endpoint ID?

blirdOP•11/7/25, 6:36 PM

Done

blirdOP•11/11/25, 10:23 AM

@Tirth

Bblird @Tirth 😄

Dj•11/11/25, 4:02 PM

Sorry, have you heard from Tirth yet? I can grab his attention or look into it myself if you need.

DDj Sorry, have you heard from Tirth yet? I can grab his attention or look into it m...

blirdOP•11/11/25, 4:04 PM

I did not yet

blirdOP•11/12/25, 9:48 AM

@Dj @Tirth we’ve spent several thousand dollars on runpod and want to continue doing so but we can’t until this is resolved

Bblird I have been trying to get the template up with multiple hf models Latest I trie...

Jason•11/12/25, 2:09 PM

is it because of your gpu choice or you select all gpus?

blirdOP•11/12/25, 2:18 PM

All gpus selected

joaoeudes7•11/12/25, 3:00 PM

I'm try using the vLlm from templante, but don't works. Only receive 500 as code of error.

joaoeudes7•11/12/25, 3:01 PM

Someone to help ?

blirdOP•11/12/25, 4:48 PM

@Jason @Tirth @Dj

joaoeudes7•11/12/25, 5:51 PM

I will request Reimbursement, plataform without support

Bblird @Jason @Tirth @Dj

Dj•11/12/25, 7:21 PM

Message me the endpoint id, I got it.

Jjoaoeudes7 I will request Reimbursement, plataform without support

Jason•11/13/25, 12:03 AM

There's always open ticket menu available on help (? Sign) , you can create a new thread or create a ticket on your own from the runpod docs (top page)

blirdOP•11/14/25, 10:01 AM

Seems like the model size of 150GB was the initial root cause. I have now deleted and redeployed with a 25GB model yet still we are stuck on Initializing with 0 workers @Tirth @Dj @Jason

Bblird I have been trying to get the template up with multiple hf models Latest I trie...

PoddyAPP•11/14/25, 10:34 AM

@blird

Escalated To Zendesk

The thread has been escalated to Zendesk!

Jason•11/14/25, 10:35 AM

Did dj told you anything?

blirdOP•11/14/25, 10:36 AM

He did tell me there was an issue with the Size of the initial model but even the smaller model causes the same problem. He said there are some PRs from the devs that might fix it but since then nothing happened

Jason•11/14/25, 10:38 AM

Ah pr takes time so you need to wait for that pr fix

Jason•11/14/25, 10:38 AM

Maybe it's on beta too, not really reliable right now I guess

blirdOP•11/14/25, 10:40 AM

Hm I see just weird that an official template doesn't work. Not saying this is anyones fault but this is business critical to use right now

Jason•11/14/25, 10:41 AM

You're using model cache right?

Jason•11/14/25, 10:41 AM

Or this is without model cache?

blirdOP•11/14/25, 10:42 AM

As far as I am concerned the vLLM template does not use it but I could be wrong

Jason•11/14/25, 10:47 AM

In your endpoint settings or did you specify a model url in the "model" section, it's not vllm worker specific

Jjoaoeudes7 Someone to help ?

Jason•11/14/25, 10:48 AM

You're experiencing the same problem or something else? Please create. Anew #

｜serverless thread if it's different

Jason•11/14/25, 10:49 AM

I thought it's either that feature (if you're using it, try to not use it for now because model cache doesn't work for you) or gpu availability but I might be wrong