Theologian Explainer (theologian-explainer-chirho)

A fine-tuned Flan-T5-base model that generates human-readable explanations of why a theological statement is classified as orthodox, heterodox, or a denominational distinctive.

Part of the Theological Guardrails Pipeline.

How It Works

Given a theological statement, its classification label, and detected heresy types, the model generates an explanation citing relevant creeds, councils, and scripture.

Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("loveJesus/theologian-explainer-chirho")
model = AutoModelForSeq2SeqLM.from_pretrained("loveJesus/theologian-explainer-chirho")

input_text = (
    "explain theological classification: Jesus is a created being, "
    "the first of God's creations. | label: heterodox | heresy types: arianism"
)

inputs = tokenizer(input_text, return_tensors="pt", max_length=512, truncation=True)
outputs = model.generate(**inputs, max_length=256, num_beams=4, early_stopping=True)

explanation = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(explanation)
# "This statement reflects Arianism, which teaches that Christ is a created being
#  rather than eternally begotten of the Father. The Council of Nicaea (325 AD)
#  condemned this view, affirming that the Son is 'begotten, not made, of one
#  Being with the Father.'"

Input Format

explain theological classification: {statement} | label: {orthodox|heterodox|denominational_distinctive} | heresy types: {comma-separated list or "none"}

Architecture

• Base model: google/flan-t5-base (248M parameters)
• Task: Conditional text generation (seq2seq)
• Max input length: 512 tokens
• Max output length: 256 tokens
• Training: Fine-tuned on ~22,500 theological statement-explanation pairs

Training Details

• Epochs: 8
• Batch size: 8
• Learning rate: 3e-5
• Optimizer: AdamW with weight decay 0.01
• Hardware: Apple M4 Pro (48GB, MPS)
• Best model selection: Lowest validation loss

Orthodoxy Basis

First six ecumenical councils (Nicaea I through Constantinople III).

Limitations

• Explanations should be verified by trained theologians
• Quality varies by heresy type and statement complexity
• Best at explaining well-known heresies with clear conciliar definitions
• May produce generic explanations for novel or ambiguous statements
• English only

Related Models

Model	Task
theologian-classifier-chirho	Multi-label heresy classification (RoBERTa-large)
theologian-embedder-chirho	Theological embeddings (MiniLM-L12)
theologian-dataset-chirho	Training dataset

License

MIT

Citation

@misc{lovejesus2026theologianexplainer,
  title={Theologian Explainer: AI-Generated Theological Classification Explanations},
  author={loveJesus},
  year={2026},
  publisher={HuggingFace},
  url={https://huggingface.co/loveJesus/theologian-explainer-chirho}
}