Cramming: Training a Language Model on a Single GPU in One Day

TL;DR

This paper explores training a transformer-based language model from scratch on a single GPU within one day, analyzing the challenges, modifications, and performance scaling laws in this highly constrained environment.

Contribution

It demonstrates that competitive language model performance can be achieved with limited compute by re-analyzing and adapting training components for a single-GPU, one-day training scenario.

Findings

Performance closely follows large-scale scaling laws

Modified training pipeline achieves results near BERT

Insights into effective modifications for limited compute training

Abstract

Recent trends in language modeling have focused on increasing performance through scaling, and have resulted in an environment where training language models is out of reach for most researchers and practitioners. While most in the community are asking how to push the limits of extreme computation, we ask the opposite question: How far can we get with a single GPU in just one day? We investigate the downstream performance achievable with a transformer-based language model trained completely from scratch with masked language modeling for a single day on a single consumer GPU. Aside from re-analyzing nearly all components of the pretraining pipeline for this scenario and providing a modified pipeline with performance close to BERT, we investigate why scaling down is hard, and which modifications actually improve performance in this scenario. We provide evidence that even in this constrained setting, performance closely follows scaling laws observed in large-compute settings. Through the lens of scaling laws, we categorize a range of recent improvements to training and architecture and discuss their merit and practical applicability (or lack thereof) for the limited compute setting.