Large Language Models, GPT-2 — Language Models are Unsupervised Multitask Learners

Large Language Models, GPT-2 — Language Models Are Unsupervised Multitask Learners

Acing GPT capabilities by turning it into a powerful multitask zero-shot model

Introduction

GPT is a well-known series of models whose last versions are currently dominating in various NLP tasks. The first GPT version was a significant milestone: being trained on enormous 120M parameters, this model demonstrated state-of-the-art performance on top benchmarks. Starting from this point, researchers tried to improve the base version.

In 2019, researchers from OpenAI officially …

capabilities gpt gpt-2 language language models large language large language models machine learning nlp series tasks transformers unsupervised versions zero-shot