Pclanglais/Epstein: A Generative LLM for English Literature

Epstein is a 13 billion parameter large language model, fine-tuned from Llama-13B, with a specialized focus on generating English literary texts. It was trained on a dataset comprising 4,000 excerpts of public domain English or English-translated literature, augmented with synthetic and manual annotations. The model's core functionality involves generating text based on up to twenty distinct literary features provided as input.

Key Capabilities

• Literary Text Generation: Creates diverse literary content based on specified parameters.
• Feature-Driven Generation: Utilizes a comprehensive set of annotations including:
  • Content: Summary, Trope, Narrative arc, Enunciation, Tone, Genre, Intertextuality, Speech standard, Literary form.
  • Context: Active/Mentioned characters, Quoted works, Absolute/Fuzzy place, Fuzzy/Absolute time, Time setting, Diegetic time.
• Companion to Brahe: Functions as the generative inverse of the analytical model Brahe, both named after characters from Daniele del Giudice's Atlante occidentale.

Good For

• Creative Writing Assistance: Ideal for authors or developers looking to generate literary drafts or explore narrative possibilities.
• Literary Experimentation: Useful for experimenting with different literary styles, genres, and narrative structures by manipulating input features.
• Educational Tools: Can be integrated into tools for studying literary elements and their impact on text generation.