11-47/high_priest_gnosticism_100k

language:

• —en license:
• —other tags:
• —gnosticism
• —high-priest
• —religious-texts
• —theology
• —nag-hammadi
• —pistis-sophia
• —spiritual
• —philosophical size_categories:
• —100K<n<1M

"Gnostic High Priest Dataset" ---

Gnostic High Priest Dataset (100k)

Dataset Description

This dataset contains 100,000 unique instruction-response pairs designed to train language models to reason and think from the perspective of a Gnostic high priest. The dataset is constructed from authentic Gnostic texts including the Nag Hammadi Library, Pistis Sophia, Gospel of Thomas, and other ancient Gnostic scriptures.

Dataset Summary

• —Total Examples: 100,000
• —File Format: JSONL (JSON Lines)
• —File Size: ~73 MB
• —Duplicates: 0 (verified unique)
• —Languages: English (with some archaic theological terminology)

Supported Tasks

• —Instruction Tuning / Supervised Fine-Tuning (SFT)
• —Teaching LLMs to adopt a Gnostic high priest persona
• —Training models to reason about Gnostic theology, cosmology, and spirituality
• —Developing context-aware religious/philosophical dialogue systems

Dataset Structure

Data Fields

Each example contains the following fields:

Type Distribution

Data Splits

Recommended split:

• —Train: 90,000 examples (90%)
• —Validation: 5,000 examples (5%)
• —Test: 5,000 examples (5%)

from datasets import load_dataset

dataset = load_dataset('json', data_files='high_priest_gnosticism_100k.jsonl')
dataset = dataset['train'].train_test_split(test_size=0.1, seed=42)
temp = dataset['test'].train_test_split(test_size=0.5, seed=42)
dataset = {
    'train': dataset['train'],
    'validation': temp['train'],
    'test': temp['test']
}

Dataset Creation

Source Data

Extracted from 22 authentic Gnostic texts:

Text Extraction and Processing

1. 1.PDF Extraction: Used PyMuPDF (fitz) to extract text from 22 PDF files
2. 2.Cleaning: Removed OCR artifacts, page numbers, headers/footers, and corrupted characters
3. 3.Chunking: Split texts into coherent chunks (200-600 characters)
4. 4.Example Generation: Created 5 types of instruction-response pairs per chunk
5. 5.Deduplication: MD5 hashing ensured zero duplicates
6. 6.Augmentation: Generated variations of instructions to reach 100k examples

Quality Control

• —Uniqueness: Verified 0 duplicates via hash checking
• —OCR Artifacts: 0 examples with excessive numbers/corrupted text
• —Response Length: Min 201, Max 805, Average 485 characters
• —Non-ASCII: Only 29 examples with special characters (intentional archaic terms)

Considerations for Use

Recommended Use Cases

• —Fine-tuning LLMs for religious/philosophical dialogue
• —Training models to understand Gnostic thought and terminology
• —Creating spiritual guidance chatbots with Gnostic perspective
• —Academic research on religious NLP

Limitations

• —Responses are based on specific Gnostic texts and may not represent all Gnostic traditions
• —Some examples contain archaic language and theological complexity
• —The "high priest" persona is a constructed interpretation, not an authentic historical voice
• —Dataset is English-only; original texts may have been translated from Coptic, Greek, etc.

Ethical Considerations

• —This dataset represents specific religious viewpoints and should be used respectfully
• —Not intended to replace authentic spiritual guidance or religious authority
• —Users should be aware that Gnostic teachings may conflict with other religious traditions
• —Models trained on this dataset should disclose their training data and perspective

Bias and Fairness

• —Dataset focuses exclusively on Gnostic Christianity and related traditions
• —Does not represent mainstream Christian, Jewish, Islamic, or other religious perspectives
• —The "high priest" framing is a specific interpretive lens

Citation

If you use this dataset, please cite:

@dataset{gnostic_high_priest_2026,
  author = {Generated for Gnostic High Priest Training},
  title = {Gnostic High Priest Dataset: 100k Instruction-Response Pairs for Training LLMs},
  year = {2026},
  howpublished = {\url{https://huggingface.co/datasets/[your-username]/gnostic-high-priest-100k}},
  note = {Dataset generated from authentic Gnostic texts including Nag Hammadi Library, Pistis Sophia, and other ancient sources}
}

Dataset Card Contact

For questions or issues regarding this dataset, please open an issue on the Hugging Face dataset repository.

Sample Data

Example 1 (priest_reflection):

{
  "instruction": "As a high priest of the Gnostic tradition, interpret this passage from 'the_gnostic_bible':\n Adam and Eve, through Seth...",
  "response": "From the sacred wisdom of the high priesthood, I contemplate this mystery: Adam and Eve were the first parents, through whom the divine light is communicated to enlightened gnostics... This reveals the path of gnosis - the divine spark within must awaken to transcend the archonic powers.",
  "type": "priest_reflection",
  "source": "the_gnostic_bible"
}

Example 2 (direct_teaching):

{
  "instruction": "Based on the Gnostic text 'Gospel-of-Philip', explain the following teaching:\n It is probably adventurous to suggest that Philip here preserves an echo...",
  "response": "It is probably adventurous to suggest that Philip here preserves an echo of Israelite reaction against the innovation, but whatever its history and antecedents the passage clearly rejects sacrifice...",
  "type": "direct_teaching",
  "source": "Gospel-of-Philip"
}

Loading the Dataset

from datasets import load_dataset

# Load from local file
dataset = load_dataset('json', data_files='high_priest_gnosticism_100k.jsonl')

# Or after uploading to Hugging Face
dataset = load_dataset('your-username/gnostic-high-priest-100k')

Training Example (with transformers)

from transformers import AutoTokenizer, AutoModelForCausalLM
from datasets import load_dataset

# Load dataset
dataset = load_dataset('json', data_files='high_priest_gnosticism_100k.jsonl')

# Format for training
def format_instruction(example):
    return {
        "text": f"### Instruction:\n{example['instruction']}\n\n### Response:\n{example['response']}"
    }

dataset = dataset.map(format_instruction)