gpt2: Building the foundations of English text generation with 124M parameters

This is the smallest version of GPT-2 released by OpenAI, an English-only language model with 124M parameters. Pre-trained on large-scale web data, the model operates by predicting the next word and provides text generation capabilities that naturally continue sentences when given a prompt.

It is compatible with major frameworks such as PyTorch, TensorFlow, and JAX, and can run on inference servers like vLLM and SGLang. With over 1,700 adapters and more than 2,200 fine-tuned models available, it is widely used as a starting point for lightweight experimentation environments or fine-tuning for specific tasks.

Trained on the Reddit-based WebText dataset excluding Wikipedia, the model is specialized for context generation rather than fact-checking. Since the model itself may reflect biases present in its training data, bias review is mandatory before deploying it in human-interactive systems.