A ~30M parameter GPT-style language model trained from scratch on financial news text.
The complete implementation, experiments, and report are available in the
GitHub repository.
The model learns the vocabulary and general structure of financial news, but generated text can contain factual and numerical errors.