Candidate models: who writes like the Book of Mormon?

One fine-tuned copy of the time-locked model per candidate author's sample (cand__...) and per control group (ctrl__...), at 125M (pilot/), 297M (base/) and 590M (mid/), as scored on the sealed Book of Mormon text in the blind run (Amendment 30). Each .pt is a state dict for the base model's architecture (load with torch.load(path, weights_only=True) into the matching time-locked model); the .json beside it is its training record and held-out scores (numbers only). alm_candidates_<size>.json is each size's validation, alm_candidates_plan.json the units. Some candidate samples come from Joseph Smith Papers transcripts, which are not published; the models are (owner's decision, 2026-11-09).

Folders: pilot/, base/, mid/ are the registered run's models. a43/<size>/ are the models retrained after Book of Mormon quotations were removed from the candidate samples (Amendment 43, post hoc; the paper's corrected rankings). a44/ holds the models of Joseph Smith's pooled non-revelatory writing (Amendment 44) and a44/a45/ the equal-dose slice models (Amendment 45), each named <unit>[_slice<k>]_<size>.pt.

The study and what it found

This is one of the models from a finished, open study of the English text of the Book of Mormon: who writes like it, what it borrows, and when its language dates from.

In short: the book copies almost nothing from earlier print. Among the eight people usually proposed as its author, the closest writing is Joseph Smith's revelations, and a search of 23,872 authors of the period puts those revelations first or second. No tool we found, ours included, can tell a person from the voice they are writing in, or reliably say "it is none of these people". The study takes no position on inspiration.

How to check the result with these

No training and no restricted text are needed.

  1. Clone the GitHub repository and download the base model of the size you want (etherfuse/timelocked-1830-125m, -300m or -590m) and the candidate files of the same size from this page (a43/pilot/ is 125M, a43/base/ 297M, a43/mid/ 590M).
  2. Load a candidate's .pt into the base model's architecture. It is a plain state dict.
  3. Score a passage: the candidate model's average loss on it, minus the base model's. The lower the number, the more that candidate's writing prepared the model for the passage. The scoring code is alm/blind_score.py.
  4. Compare with ours: the scores we got for every Book of Mormon segment are in data/blind/ in the GitHub repository.

About

Part of an open, pre-registered study of the authorship, borrowing and dating of the earliest Book of Mormon text (https://github.com/lifehug/book-of-mormon-authorship; pre-registration https://osf.io/6ugxc/). The project does not argue for or against divine inspiration. Everything is published so it can be checked and redone; the corpus texts we agreed to keep private are not, and the repository's docs/replication.md gives the steps to obtain the same access.

Licence: CC BY-NC 4.0. Training data: English printed 1780-1830 (Internet Archive, Harvard Institutional Books 1.0, HathiTrust public-domain volumes, Text Creation Partnership), with the Book of Mormon and everything after 1830 removed; see the base model cards (etherfuse/timelocked-1830-125m). Contains derivatives of Harvard Institutional Books 1.0, Institutional Data Initiative, Harvard Law School Library; used for noncommercial research. Checksums: every file's SHA-256 is in the GitHub repository, data/manifests/hf_<repo>.jsonl.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support