How to Train BERT with an Academic Budget
This enables researchers with limited budgets to train BERT-like models, though it is incremental as it optimizes existing methods rather than introducing new paradigms.
The paper tackles the problem of high computational cost for pretraining BERT models by presenting a recipe to train a masked language model in 24 hours on a single low-end server, achieving competitive performance with BERT-base on GLUE tasks at a fraction of the original cost.
While large language models a la BERT are used ubiquitously in NLP, pretraining them is considered a luxury that only a few well-funded industry labs can afford. How can one train such models with a more modest budget? We present a recipe for pretraining a masked language model in 24 hours using a single low-end deep learning server. We demonstrate that through a combination of software optimizations, design choices, and hyperparameter tuning, it is possible to produce models that are competitive with BERT-base on GLUE tasks at a fraction of the original pretraining cost.