Skip to content

Research

[Paper] Graph Machine: Towards Better Pretraining via Edges

GM-2: Ultra-sparse 2–4/4096 dynamic routing replaces most Qwen3-0.6B layers, preserving pretraining loss over 15B tokens.

Read more →

2026-09-02

[Blog] Graph Machine: Intuitions

Two intuitions for Graph Machine (2), one algorithmic and one biological.

Read more →

2026-08-24

[Paper] Graph Machine: Exploring Edge Mechanisms as an Inductive Bias

GM-1: Investigating whether explicit, dynamically updatable edges improve relational reasoning on Sudoku.

Read more →

2026-08-07