A sentence transformer fine-tuned for biblical verse similarity and cross-reference discovery. Given a verse text, it produces a 384-dimensional embedding that places semantically related verses close together in vector space.

Training

Base model: sentence-transformers/all-MiniLM-L12-v2
Loss: Triplet loss (cosine distance, margin=0.5)
Data: 344,798 triplets from the Treasury of Scripture Knowledge (OpenBible.info)
- Anchor: verse A, Positive: cross-referenced verse B, Negative: hard negative (same-book unrelated verse)
Epochs: 3
Batch size: 64
Device: Apple MPS (M4 Pro)

Evaluation

Triplet ranking accuracy: 86.75% (positive cross-ref ranked higher than negative)
Separation gap: 0.4213
Pearson cosine: 0.6271
Spearman cosine: 0.6326

Usage

python

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("LoveJesus/intertextual-embedder-chirho")

verses = [
    "In the beginning God created the heaven and the earth.",
    "In the beginning was the Word, and the Word was with God, and the Word was God.",
    "And the children of Israel went into the midst of the sea upon the dry ground.",
]

embeddings = model.encode(verses)
# embeddings[0] will be closest to embeddings[1] (Gen 1:1 <-> John 1:1)

Part of models-chirho

This model is part of the Intertextual Reference Network pipeline, paired with:

Classifier: LoveJesus/intertextual-classifier-chirho
Dataset: LoveJesus/intertextual-dataset-chirho

Built with love for Jesus by loveJesus.