Vision Model update

1 replies
0 attachments
Started >30d ago
[AutoMod] action=keep confidence=0.99 | OP announcing technical update with technical details and call for feedback

We have updated our internal infrastructure for our NSFW vision model. Cyberix now uses an alternative stack that handles non-photographic content (drawings, anime, illustrations) more accurately than the previous system.

It also carries a lighter resource footprint and processes attachments faster, which lets us direct more of the server toward the other services we host.

Let us know if you encounter any issues with false positives/negatives.
Our new stack partially takes from the Wingman Jr project:
https://github.com/wingman-jr-addon/wingman_jr

[AutoMod] action=keep R:8 E:5 N:3 C:10 | Partially relevant to the thread's discussion of infrastructure changes and trade-offs, but leans into skepticism rather than constructive engagement. Includes a mild critique of the update but does not provide new data or alternative perspectives.

Oh, so long as you're not trying to scan some sketchy selfie or family pics then I guess this is less of a doomsday scenario. Still think we should be keeping the old stack around for emergency runs.

[CN]

Reply

Posting anonymously. Your IP address will be recorded for rate limiting purposes.





Max 10MB per file. Allowed: images, videos, audio, PDF, text, zip