Per un attimo Brahe cercò le parole, le immagini, le analogie; pensò perfino i gesti della mano e delle dita, come un attore che si prepari a rendere fisico un sentimento. Ma appena cominciò a dire "come", a dare solidità a ciò che non aveva, a rendere visibile ciò che non lo era, a collocare, nello spazio ciò che era pura probabilità, e a cercare una qualsiasi cosa tra le forme del mondo cui paragonarlo, Epstein lo interruppe.
Daniele del Giudice, Atlante occidentale

Epstein is a generative LLM for English literature fine-tuned from llama-13B. Given twenty potential features, Epstein will generate a literary text.

Epstein has been trained on 4,000 excerpts of English or English translated literature in the public domain and on a set of synthetic and manual annotations.

Epstein is the reversed companion of Brahe, an analytical LLM model to analyze existing texts using the same features. Both models are named after the protagonists of the philosophical novel of Daniele del Giudice, Atlante occidentale. Brahe is a scientist working at the CERN on quantum physics, Epstein is a novelist and they both confront their different views of reality.

Annotations

In its current version, Epstein can generate texts using any of the following annotations. It's preferable to include at least a summary.

  • Summary: short summary
  • Trope: a trope or literary cliché (a fuzzy definition but works surprisingly well)
  • Narrative arc: how is the action unfolding (suspense, dramatic tension, comic relief…)
  • Enunciation: who is speaking in the text (first-person narrative, dialog, third-person narrative, omniscient narrator)
  • Tone: general tonality of the text (humoristic, tragic, scholarly…)
  • Genre: a specific literary genre that would be used in bookshops such as detective fiction, science-fiction, romance, historical novel, young adult…
  • Intertextuality: non-literary writing forms that may be similar to this text (red tape, scientific article, case law…)
  • Speech standard: the specific social/literary level of the text (poetic, dialectical, vulgar…)
  • Literary form: whether it's the description of a place, a conversation, a stream of consciousness
  • Literary movement: aesthetic movement the text seems to embody (does not work so well)
  • Active character: the list of characters that have an active involvment in the story.
  • Mentioned characters: the list of characters only mentioned, with no active involvement in the story
  • Quoted works: another text mentioned or quoted in the text.
  • Absolute place: a precise place with a proper name such as Paris, Sesame Street, Lisbonne Airport.
  • Fuzzy place: unnamed place where the story happens such as a field, an appartment, a church (does not work so well…)
  • Fuzzy time nonspecific moment where the action occur moment such as monday, yesterday, a week after.
  • Time setting: historical period where the action seems to occur such as the 1960s, the Renaissance, the Victorian period…
  • Diegetic time: very approximative number of minutes/hours/days that have unfolded between the beginning and the end of the text (5 minutes, 35 minutes, 2 hours, 3 days).
  • Absolute time: a precise date where the action occurs, such as January 15, 1845, 23rd century…

Example

A science-fiction rewriting of the Ambassadors of Henry James. The prompt uses the same features created by Brahe with a few twists:

Summary: Strether is waiting for his friend Waymarsh to arrive at the hotel, but is not completely disappointed when he learns that Waymarsh will not arrive until evening.
Trope: The anticipation of meeting a friend
Narrative arc: Suspense
Enunciation: Third-person narrative
Tone: Anticipatory
Genre: Science-fiction
Intertextuality: Travelogue
Speech standard: Standard
Literary form: Stream of consciousness
Literary movement: Realism
Active character: Strether, Waymarsh
Time setting: the year 2046
Diegetic time: A few hours
Fuzzy place: The hotel

Output of Epstein:

Strether waited for him at the hotel, but when he learned that his friend would not arrive until evening, he was not completely disappointed. He had plenty to do in the meantime, and he felt a certain relief at having a few days to prepare for the meeting. He had been looking forward to it with a mixture of anticipation and dread, and he was now glad to have some time to think about it and to plan how they would spend their time together. The year 2046 was a time of great change and progress. Strether had been living in the city of New York, which had become a hub of innovation and creativity. The city was home to many of the world's leading scientists and engineers, who were working on cutting-edge projects that were changing the way people lived and worked. Strether had been working as a researcher himself, and he had been excited to learn that Waymarsh was coming to visit him. Waymarsh was a friend from college, and they had stayed in touch over the years. They had both been interested in science and technology, and they had shared a love of exploring new ideas and discoveries. Strether was looking forward to catching up with Waymarsh and hearing about his latest projects.

Downloads last month
13
Safetensors
Model size
13B params
Tensor type
BF16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.