Search papers, labs, and topics across Lattice.
1
0
3
Training VLMs on context lengths matching evaluation lengths yields better performance than training on even longer contexts, challenging common assumptions about scaling laws.
