Engram Gyre

--- license: apache-2.0 library_name: transformers pipeline_tag: text-generation datasets: - TrueV1sion123/engram-gyre-behavior language: - en tags: - engram - gyre - recursive-abstraction - cognition - reasoning-structure - llama - from-scratch --- # engram-gyre **The first model in the engram lineage with actual trained weights.** A ~11M-parameter, from-scratch, Llama-architecture language model whose entire behavioral repertoire is a single **gyre revolution**: the four-phase Recursive Abstraction Engine loop treated as a *spiral*, executed with the cognitive moves of the [`engram-prime`](https://huggingface.co/TrueV1sion123) MINDSKILL. Give it a problem; it performs one revolution: ``` <SATURATION> immerse in the interior of the problem, past the surface <ABSTRACTION> extract the irreducible axiom + a cross-domain isomorph (shorter than saturation) <DESCENT> bring the abstraction down to concrete, testable consequences <INTEGRATION> compress to a synthesis that cannot be misread (+ honest <CLICK> mark) <REENTRY> the gyre turns — the higher-level question the answer opens, + honest residue ``` > A gyre is not a circle. A circle returns to where it began; a gyre returns to the same angle > one turn higher. The question you re-enter is the one you left, displaced upward by exactly > the understanding the last revolution deposited. This model completes the [`rae-training`](https://huggingface.co/TrueV1sion123/rae-training) repository, which shipped the methodology, loss design, and theory but **no weights**. `engram-gyre` is the weights. > ### Headline finding — does the gyre form improve reasoning? (controlled study) > We ran the study the whole project pointed at: 3 arms on Qwen2.5-3B-Instruct — **gyre vs plain > chain-of-thought vs direct-answer**, identical content, only the reasoning *format* differs — > judged blind on answer quality over 35 held-out problems ([`docs/REASONING_STUDY.md`](./docs/REASONING_STUDY.md)). > **Result:** reasoning of either kind *massively* beats direct answering (gyre & CoT ≈ 24/30 vs > direct 17/30), but the **gyre form gives no answer-quality advantage over plain CoT** — CoT edges > it (55% vs 3% on blind pairwise; gyre never scored higher on any single problem; sign-test p=0.03). > The honest conclusion: **the gyre form is a legibility-and-discipline scaffold — structured, > compressed, auditable reasoning — not...