The Entropy Wall

Conventional exact compression of trained models stops at a limit, and the limit is shared across the model families we measured.

Kind
Paper
Programme
FCN Compression
Published
27 June 2026
Status
Published

What we found

We measured trained models from several families, as they are commonly stored and shared. In each, conventional exact compression reached the same limit. In the models we measured, the limit came from the models themselves, not from a weakness in any one compressor.

Where the Forlais Theorem sets the limit in principle, the Entropy Wall is where it was met in practice.

Why it matters

A limit shared by every model we measured explains why established compressors gained little on them, and why going further calls for a different starting point rather than a better compressor.

The paper

BF16 1.52x Entropy Ceiling

Oliver Cooper, Forlais Group. Published 27 June 2026.

Read the paperPDF, 513 KBDOI 10.5281/zenodo.20961807CC BY 4.0

Cite this record

Forlais Research (2026). The Entropy Wall. Forlais Research Library, record FR-F-02.

@misc{fr-f-02,
  author = {{Forlais Research}},
  title = {{The Entropy Wall}},
  howpublished = {Forlais Research Library, record FR-F-02},
  year = {2026},
  note = {27 June 2026}
}