GLaM: Efficient Scaling of Language Models with Mixture-of-Experts
This work addresses the efficiency problem for training and deploying large-scale language models, offering a more sustainable and cost-effective approach.
The paper tackles the high computational cost of scaling large language models by proposing GLaM, a sparsely activated mixture-of-experts architecture, which reduces training energy by 2/3 and inference flops by half compared to GPT-3 while achieving better performance on 29 NLP tasks.
Scaling language models with more data, compute and parameters has driven significant progress in natural language processing. For example, thanks to scaling, GPT-3 was able to achieve strong results on in-context learning tasks. However, training these large dense models requires significant amounts of computing resources. In this paper, we propose and develop a family of language models named GLaM (Generalist Language Model), which uses a sparsely activated mixture-of-experts architecture to scale the model capacity while also incurring substantially less training cost compared to dense variants. The largest GLaM has 1.2 trillion parameters, which is approximately 7x larger than GPT-3. It consumes only 1/3 of the energy used to train GPT-3 and requires half of the computation flops for inference, while still achieving better overall zero-shot and one-shot performance across 29 NLP tasks.