Search papers, labs, and topics across Lattice.
Affiliation:
2
0
4
Most directions in a transformer's residual stream are functionally invisible to the terminal loss, exposing a thin observable subspace whose pullback Fisher geometry consistently outperforms standard attention weights for compression and distillation.
Drifting Models get a friction boost: DMF matches or beats Optimal Flow Matching on FFHQ image translation while slashing training compute by 16x.
