Changed to a longer tag name to prevent conflicts with existing repos

#16
by Steveeeeeeen HF Staff - opened

Sorry, I realized that some non related models has this tag. I've changed to a more explicit one!

Nari Labs org

Thank you!

NariLabs changed pull request status to merged

kudos on the great model @NariLabs !

Hi! I took a look at the public endpoints for the Space. Whenever I queue a request (/gradio_api/call/generate_audio) I get an event_id, but polling /gradio_api/result/<id> always returns 404 and /queue/status currently responds with 500 Internal Error. That points to the worker crashing right after the job is enqueued.

Because I don’t have maintainer access, I can’t retrieve the runtime logs from the CLI, so could someone who does run hf spaces logs nari-labs/Dia-1.6B --limit … (or check the Space dashboard) to grab the stack trace?

One thing that stands out is the Space metadata: the front-matter reads python_version: 3.10, but the Hub seems to have parsed it as the float 3.1. That can put the build in a broken environment. I’d recommend changing it to a string (python_version: "3.10") and redeploying.

Once that’s fixed, also double-check that Dia.from_pretrained("nari-labs/Dia-1.6B-0626") can actually download the weights and that descript-audio-codec (via dac.utils.download()) is allowed during startup—failures there are another common cause of the worker exiting.

Hope this helps! Let me know if you need more detail.

Sign up or log in to comment