Reducing the High Cost of Training NLP Models With SRU++

The increasing computation time and costs of training natural language models (NLP) highlight the importance of inventing computationally efficient models that retain top modeling power with reduced or accelerated computation. A single experiment training a top-performing language model on the 'Billion Word' benchmark would take 384 GPU days and as much as $36,000 using AWS on-demand instances.

Check out the full article at KDNuggets.com website
Reducing the High Cost of Training NLP Models With SRU++