Learn Before
Classification

Three Phases of LLM Scaling with Dataset Size

The relationship between a Large Language Model's test error and the size of its training dataset can be characterized by three distinct stages when viewed on a log-log plot. The process begins with a 'Slow Reduction Phase,' transitions into a 'Power-law Reduction Phase' of rapid improvement, and concludes with a 'Convergence Phase' where performance gains level off as they approach an irreducible error.

0

1

Updated 2026-04-21

Contributors are:

Who are from:

Tags

Ch.2 Generative Models - Foundations of Large Language Models

Foundations of Large Language Models

Foundations of Large Language Models Course

Computing Sciences