recursive-reasoning

CGAR: curriculum-guided adaptive recursion. 1.71x faster training for tiny recursive reasoning models (10.93h to 6.38h on 423k Sudoku-Extreme puzzles)

Benchmark for evaluating advanced reasoning, recursive dependency resolution, and robustness capabilities of large language models in dynamic, noisy, and structurally challenging environments.

Disentangling gradient quality from architecture in recursive reasoning. Controlled experiment: 1-step gradient approximation is the sole bottleneck in HRM vs TRM performance gap.

Completing the 2x2 factorial: HRM's hierarchy with full BPTT reveals gradient-architecture interaction

"TRM (Tiny Recursive Model) integration architecture for Symbion.space ecosystem"

Self-evolving recursive AI reasoning system — φ-governed tool use, multi-agent code review, workflow orchestration, and persistent spiral memory, built on E8 Lie group geometry and golden ratio mathematics

Unlock infinite context for LLMs. Transforms language models into orchestrators using code generation and parallel delegation for complex tasks beyond token limits.

TinyRecurseLab — train a real Tiny Recursive Model (TRM) in your browser on 4x4 Sudoku. From-scratch autograd, 6 interactive modules on recursive reasoning (arXiv:2510.04871).

NL-CGAR: routing natural-language problems to curriculum-guided recursive reasoning experts. Paper under review at Expert Systems with Applications

First 1-bit (BitNet b1.58) recursive reasoner for Sudoku-Extreme - distilled from a 7M-param FP TRM teacher into a 1.4 MB ternary student

Studying Tiny Recursive Models (TRM) and recursive reasoning for abstract reasoning.

Tiny Recursive Model (TRM) for grid-based local navigation — neural pathfinding that imitates A* via recursive reasoning