A Tilted Bowl Is Not a Slippery Slope: Compressing Looped Models

Excerpt

arXiv:2609.39277v1 Announce Type: new Abstract: Looped models reason by applying the same block of weights many times, so compressing that block saves memory traffic on every loop. Compressed looped models, however, often collapse, and the collapse is usually blamed on rounding error that accumulates from loop to loop. In this work we test that account on more than 30 models from five families and find, to our surprise, that it holds only for loops that never settle. When a loop settles, a fixed