Reposted by benjamin
I'll get straight to the point.
We trained 2 new models. Like BERT, but modern. ModernBERT.
Not some hypey GenAI thing, but a proper workhorse model, for retrieval, classification, etc. Real practical stuff.
It's much faster, more accurate, longer context, and more useful. 🧵
![White-on-black text saying "In fact, [MASK]-large’s processing speed is closer to that of [MASK]-base than it is to [MASK]-large's.", with the [MASK] drawn in purple toi draw attention](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:fjfvhasy573zvbixh4ynp6g4/bafkreianson5ll3onihxnouxfe5w2fxogojczuijs76tz54qpjhsrvhx7m)
