Search papers, labs, and topics across Lattice.
1
0
3
2
You can slash ModernBERT's pre-training data by over 90% and still match performance, simply by prioritizing dataset diversity.
