Meta Releases 175B Parameter Language Model to Advance AI Research
Meta AI shares its massive OPT-175B language model with researchers to democratize access and improve understanding of AI risks. The release includes both the pretrained model and the underlying code under a noncommercial license.
Meta Platforms Inc. announces that its artificial intelligence research team shares the Open Pretrained Transformer (OPT-175B) model with the scientific and academic research community. This large language model contains 175 billion parameters and trains on publicly available data sets to perform tasks like creative text generation, basic math solving, and reading comprehension.
The company takes this step to democratize access to advanced natural language processing systems that previously remain locked behind paid APIs and limited to a few highly resourced labs. By providing both the pretrained models and the necessary code, Meta AI allows researchers to investigate how these massive algorithms work and tackle ongoing issues such as bias and toxicity.
Meta AI releases OPT-175B under a noncommercial license that restricts the technology to research use cases only. The team hopes this unprecedented level of transparency brings more voices to the forefront of large language model creation and helps the community design responsible strategies for mitigating potential harms.