Train-to-Test (T2) Scaling Laws

Train-to-Test (T2) scaling laws is a research framework developed by academics at the University of Wisconsin-Madison and Stanford University. It addresses the disconnect between traditional LLM pretraining scaling laws (optimizing for training cost) and real-world inference needs (where techniques like repeated sampling increase accuracy but also cost). T2 jointly optimizes a model's parameter size, training data volume, and the number of test-time inference samples to achieve compute-optimal performance for reasoning-heavy applications.