← All works
Core recurrence

Compressing Looped Models

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

Steven Kolawole, Pearse Jim, Opegbemi M. Busoye, Glory Bagai, Virginia Smith

Tests the common claim that compressed looped models collapse because rounding error accumulates across loops. Across more than 30 models from five families, this holds only for loops that never settle; for settling loops, error shifts the fixed point instead of accumulating.

Inside the method

Low-bit [ Block ] × r → halting head fires → 8-bit [ Block ] × few

Simplified conceptual schematic. Consult the paper for the complete architecture.

Recurrence family
Core recurrence
Depth control
Halting-head controller plus final high-precision loops
KV / state strategy
See paper

Reading note

A single label-free measurement predicts which compressed models fail. The controller beats fixed-depth inference by up to 15 points on Sudoku-Extreme and Maze-Hard with under a third of the weight traffic.

Sources checked 2026-10-28. This catalog does not imply independent reproduction.

Related works

Cite this work

@misc{tiltedbowlcompression2026,
  title = {A Tilted Bowl Is Not a Slippery Slope: Compressing Looped Models},
  author = {Steven Kolawole and Pearse Jim and Opegbemi M. Busoye and Glory Bagai and Virginia Smith},
  year = {2026},
  eprint = {2609.39277},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.39277}
}

Suggest a correction on GitHub ↗