GPT-2
GPT-2 is a 1.5 billion parameter language model released by OpenAI in a staged rollout that served as a test case for responsible AI model publication. Since its 2019 release, GPT-2 has become a widely-used baseline for evaluating newer models and architectures, with recent research using it for applications ranging from aspect-based sentiment analysis to code generation (CodeParrot) and neuron interpretability studies. The model continues to be optimized across various hardware platforms including ONNX Runtime and Graphcore IPU processors, and remains a reference point for architectural improvements in the transformer-based model landscape.
Updated 3 August 2026
Specifications
No specifications recorded yet.
Latest developments
Meet Gigatoken: A Rust BPE Tokenizer that Encodes Text at 24.53 GB/s, up to 989x Faster than HuggingFace Tokenizers
MarkTechPost · 1 week ago ·
32
CoFrGeNets replace the ‘bones’ of transformer-based models
IBM Research · 3 weeks ago ·
13
From GPT-2 to gpt-oss: Analyzing the Architectural Advances
Ahead of AI · 11 months ago ·
30
Data Machina #257
Data Machina · 2 years ago ·
8
SetFitABSA: Few-Shot Aspect Based Sentiment Analysis using SetFit
Hugging Face Blog · 2 years ago ·
11
Accelerating over 130,000 Hugging Face models with ONNX Runtime
Hugging Face Blog · 2 years ago ·
15
Language models can explain neurons in language models
OpenAI Blog · 3 years ago ·
30
Graphcore and Hugging Face Launch New Lineup of IPU-Ready Transformers
Hugging Face Blog · 4 years ago ·
1
July 2026
- Meet Gigatoken: A Rust BPE Tokenizer that Encodes Text at 24.53 GB/s, up to 989x Faster than HuggingFace Tokenizers
- CoFrGeNets replace the ‘bones’ of transformer-based models
August 2025
June 2024
December 2023
October 2023
May 2023
May 2022
December 2021
January 2021
November 2019
September 2019
August 2019
April 2019
January 2019
Relationships
Products & technology
- OpenAI develops this model · 4 sources
- Graphcore integrated with this model · 1 source
- Hugging Face integrated with this model · 1 source
- CodeParrot derived from this model · 1 source
- MuseNet derived from this model · 1 source