Qwen3-Omni sits through 93,012 tracks for auto-captioning, dies on track 8045 with a JSONDecodeError

tetsuoai · x · 2026-09-23

A user jokes about asking "are we there yet" every 90 seconds while Qwen3-Omni listens to track 8045 of 93,012 in its entirety to generate captions — only to crash mid-caption with a JSONDecodeError. A relatable snapshot of large-scale audio annotation runs going sideways.

Related event: Qwen3-Omni Patiently Listens Through 93,012 Songs Until JSON Crash(2 posts)→

Original post →

More from Fun

Fun channel →