Organic chemistry

Key (lock)

Natural language understanding

Greenhouse gas

Memory footprint

Biology

Machine learning

Engineering

Footprint

Language model

Paleontology

Artificial intelligence

Natural language

Systems engineering

Chemistry

Ecology

Programming language

Natural language processing

Shot (pellet)

Computer science

Task (project management)

Carbon footprint

Computer security

When scaled to hundreds of billions of parameters, pretrained language models such as GPT-3 (Brown et al., 2020) achieve remarkable few-shot performance. However, enormous amounts of compute are required for training and applying such big models, resulting in a large carbon footprint and making it difficult for researchers and practitioners to use them. We show that performance similar to GPT-3 can be obtained with language models that are much “greener” in that their parameter count is several orders of magnitude smaller. This is achieved by converting textual inputs into cloze questions that contain a task description, combined with gradient-based optimization; exploiting unlabeled data gives further improvements. We identify key factors required for successful natural language understanding with small language models.

It’s Not Just Size That Matters: Small Language Models Are Also Few-Shot Learners