GPT-2 (1.5B)
OpenAI · 2019-11-05 · 1.6B parameters
GPT-2 (1.5B) is the largest variant of OpenAI's second-generation language model, released in November 2019 with 1.5 billion parameters. It is a decoder-only transformer trained on a large web-scraped text corpus (WebText) with a next-token prediction objective, notable at the time for its surprisingly fluent long-form text generation. GPT-2 was staged in its release due to misuse concerns and remains an important research baseline for language modeling.
Benchmark scores
| Benchmark | Score |
|---|---|
| Capability | 63.5 |