GPT-Neo

E435863

GPT-Neo is an open-source family of autoregressive language models developed by EleutherAI as a free alternative to OpenAI’s GPT-3.

All labels observed (1)

Label Occurrences
GPT-Neo canonical 1

How this entity was disambiguated

Statements (49)

Predicate Object
instanceOf autoregressive language model family
language model
language model
language model
open-source language model
accessModel downloadable checkpoints
self-hosted
architectureType decoder-only transformer
basedOn transformer architecture
compatibleWith Hugging Face Transformers NERFINISHED
designedAs free alternative to GPT-3
developer EleutherAI NERFINISHED
framework PyTorch NERFINISHED
TensorFlow NERFINISHED
hasCommunity EleutherAI community
hasModel GPT-Neo 1.3B NERFINISHED
GPT-Neo 125M NERFINISHED
GPT-Neo 2.7B NERFINISHED
hostedOn GitHub NERFINISHED
inspiredBy GPT-3 NERFINISHED
language English
license MIT License
openSource true
parameterCount 1.3 billion
125 million
2.7 billion
programmingLanguage Python
relatedTo GPT-J NERFINISHED
GPT-NeoX-20B NERFINISHED
releasedBy EleutherAI NERFINISHED
releaseYear 2021
repositoryName EleutherAI/gpt-neo NERFINISHED
supportsInference CPU
GPU NERFINISHED
supportsTask conditional text generation
few-shot learning
language modeling
text completion
text continuation
text generation
zero-shot learning
tokenizerType byte pair encoding
trainingData The Pile NERFINISHED
trainingDataDeveloper EleutherAI NERFINISHED
trainingObjective next token prediction
trainingType unsupervised learning
useCase application integration
prototyping
research

How these facts were elicited

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.