Language models (LMs) have become ubiquitous in both NLP research and incommercial product offerings. As their commercial importance has surged, themost powerful models have become closed off, gated behind proprietaryinterfaces, with important details of their training data, architectures, anddevelopment undisclosed. Given the importance of these details inscientifically studying these models, including their biases and potentialrisks, we believe it is essential for the research community to have access topowerful, truly open LMs. To this end, this technical report details the firstrelease of OLMo, a state-of-the-art, truly Open Language Model and itsframework to build and study the science of language modeling. Unlike mostprior efforts that have only released model weights and inference code, werelease OLMo and the whole framework, including training data and training andevaluation code. We hope this release will empower and strengthen the openresearch community and inspire a new wave of innovation.
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)