Just wanted to leave some feedback on the “recently created” feed. It’s become almost impossible to use for discovering new models because it’s constantly flooded with auto-generated repos. It’s not just random names, but programmatic ones that are clearly from scripts.
Here’s a quick screenshot of what it looks like right now
It feels like this could be managed with a pretty simple technical solution. What about a small, fast classifier like a finetuned DistilBERT to check the repo name on creation? An in-house model could also work.
If the name looks like spam or is clearly robotic, just show a warning and add a short 10-15 second cooldown on the ‘create’ button, maybe a CAPTCHA. It’s a small bit of friction that would massively improve the signal-to-noise ratio on the Hub without blocking legitimate users.