GPT-NeoX-20B: An Open-Source Autoregressive Language Model

TL;DR

GPT-NeoX-20B is a large, open-source 20-billion-parameter autoregressive language model trained on the Pile, demonstrating strong few-shot reasoning capabilities and outperforming comparable models in various tasks.

Contribution

This work introduces GPT-NeoX-20B, the largest openly available dense autoregressive model, with detailed architecture, training methodology, and performance evaluation.

Findings

GPT-NeoX-20B excels as a few-shot reasoner.

It outperforms similar-sized GPT-3 and FairSeq models in five-shot evaluations.

The model and training code are openly available for research.

Abstract

We introduce GPT-NeoX-20B, a 20 billion parameter autoregressive language model trained on the Pile, whose weights will be made freely and openly available to the public through a permissive license. It is, to the best of our knowledge, the largest dense autoregressive model that has publicly available weights at the time of submission. In this work, we describe \model{}'s architecture and training and evaluate its performance on a range of language-understanding, mathematics, and knowledge-based tasks. We find that GPT-NeoX-20B is a particularly powerful few-shot reasoner and gains far more in performance when evaluated five-shot than similarly sized GPT-3 and FairSeq models. We open-source the training and evaluation code, as well as the model weights, at https://github.com/EleutherAI/gpt-neox.