Sparse Model Nexus (LLM↔Quantum continuum)

A Pharosiraptor knowledge-graph publication · snapshot 2026-07-06

Scope & honesty note. This is a proof of mechanism, not a competitive language model. Three LoRA adapters are actually trained (distilgpt2, a handful of sentences each); their demo generations are incoherent by design at this scale. The ~150 other graph nodes are a proposed type inventory — Pharosiraptor ops catalogued as candidate composable units — not trained models. Framework terms are borrowed loosely: “Tarski residual” denotes irreducible uncertainty + distribution shift, not a consequence of Tarski’s theorem; “maximal self-organized criticality” is a proposed, unmet requirement — the one composition we measured contradicts it. Full self-critique →

WHAT THIS IS

One sparsity dial spans dense LLMs to the sparse-limit model — the minimal composable unit (a LoRA adapter, called a “quantum” here). Quanta compose via a four-operator algebra (⊕ mixture, ∘ sequential, ▷ route, ⊗ merge), selected online by a cost-aware receding-horizon planner over forecast workloads. A proposed non-redundancy requirement (“maximal criticality”) is not yet met; the planner’s leftover uncertainty is named — loosely — a “Tarski residual.” Read the scope note above before the vocabulary.

START HERE — routed by the SMN planner (applied to this site)

Pick your role; the planner composed the minimal reading path that covers what you need. How this was computed →

Board / exec	`▷ route`	Executive view	1 pg · 518 words
Engineer	`▷ route`	Operator demo + model	1 pg · 715 words
Researcher	`▷ route`	Executive view	1 pg · 518 words
Newcomer	`▷ route`	Executive view	1 pg · 518 words
Skeptic	`▷ route`	Executive view	1 pg · 518 words
Due-diligence	`⊕ mixture×2`	Executive view + Operator demo + model	2 pg · 1,233 words

Sparse Model Nexus (LLM↔Quantum continuum)

Executive view

Operational view

Operational demo & model ↓

Realized future state

Research note & evaluation