CGAR: curriculum-guided adaptive recursion. 1.71x faster training for tiny recursive reasoning models (10.93h to 6.38h on 423k Sudoku-Extreme puzzles)
-
Updated
Aug 12, 2026 - Python
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.
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
NL-CGAR: routing natural-language problems to curriculum-guided recursive reasoning experts. Paper under review at Expert Systems with Applications
Recursive reasoning skills for LLM harnesses, including Multipath Reasoning and recursive confidence-loop scoring.
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
Tiny Recursive Model (TRM) for grid-based local navigation — neural pathfinding that imitates A* via recursive reasoning
To associate your repository with the recursive-reasoning topic, visit your repo's landing page and select "manage topics."