Ellesia - Seventh-day Adventist AI Assistant

Model Description

Ellesia is a fine-tuned large language model designed to provide biblically grounded answers aligned with Seventh-day Adventist theology. Built on GPT-OSS-120B and fine-tuned using LoRA (Low-Rank Adaptation), Ellesia is trained on 3,875 conversations covering SDA doctrine, Scripture interpretation, Ellen G. White's writings, and pastoral guidance.

Model Details

  • Base Model: GPT-OSS-120B (120 billion parameters)
  • Fine-tuning Method: LoRA (Low-Rank Adaptation)
  • Training Data: 3,875 SDA theological conversations (Dataset)
  • Training Framework: Tinker API
  • Training Loss: 9.013 → 0.140 (98.5% improvement)
  • LoRA Rank: 32
  • Epochs: 12
  • License: Apache 2.0

Capabilities

  • Biblically Grounded Responses: Provides answers based on Scripture and SDA theological principles
  • Spirit of Prophecy Integration: References Ellen G. White's writings where appropriate
  • Doctrinal Accuracy: Aligned with the 28 Fundamental Beliefs of Seventh-day Adventists
  • Pastoral Guidance: Offers practical spiritual counsel and ministry support
  • Biblical Interpretation: Explains Scripture within the Adventist hermeneutical framework

Usage

Using with Tinker API

import tinker
import asyncio

async def chat_with_ellesia():
    # Initialize Tinker client
    service_client = tinker.ServiceClient(api_key="your-tinker-api-key")

    # Load Ellesia model
    sampling_client = service_client.create_sampling_client(
        model_path="tinker://b93c0d5c-9dbb-4156-bfeb-a5e8896732a5/sampler_weights/final"
    )

    # Prepare conversation
    messages = [
        {
            "role": "system",
            "content": "You are a knowledgeable Seventh-day Adventist church assistant."
        },
        {
            "role": "user",
            "content": "What does the Bible teach about the Sabbath?"
        }
    ]

    # Generate response
    response = await sampling_client.sample_async(
        messages=messages,
        max_tokens=500,
        temperature=0.7
    )

    print(response)

asyncio.run(chat_with_ellesia())

Interactive Chat

Clone the repository and use the interactive chat interface:

git clone https://github.com/imaiimaiimaiimaiimaiimai/Ellesia.git
cd Ellesia
pip install -e .

# Set your Tinker API key
export TINKER_API_KEY="your-api-key-here"

# Start chatting
./run_ellesia.sh

Training Details

Training Data

The model was fine-tuned on the Seventh-day Adventist Conversations Dataset, which contains 3,875 high-quality conversations covering:

  • SDA doctrine and the 28 Fundamental Beliefs
  • Biblical interpretation and hermeneutics
  • Ellen G. White's writings and the Spirit of Prophecy
  • Church history and denominational heritage
  • Practical Christian living and pastoral care
  • Sabbath theology and observance
  • Prophetic interpretation (Daniel, Revelation)
  • Health message and lifestyle

Training Hyperparameters

  • Base Model: openai/gpt-oss-120b
  • Method: LoRA (Low-Rank Adaptation)
  • LoRA Rank: 32
  • Learning Rate: 1e-4
  • Batch Size: 256
  • Epochs: 12
  • Max Sequence Length: 8192 tokens
  • Optimizer: AdamW
  • Loss Function: Cross-entropy

Training Results

  • Initial Loss: 9.013
  • Final Loss: 0.140
  • Improvement: 98.5%

Limitations and Responsible Use

Theological Perspective

  • This model reflects Seventh-day Adventist theological interpretations
  • Should be used as a supportive tool, not a replacement for pastoral counsel
  • Biblical references should be verified in Scripture

Known Limitations

  • May occasionally require fact-checking against official SDA sources
  • Complex theological questions may benefit from consultation with church leadership
  • Not intended for medical, legal, or financial advice

Model Sources

Citation

@misc{ellesia2024,
  title={Ellesia: A Fine-tuned AI Assistant for Seventh-day Adventist Ministry},
  author={Ellesia Project},
  year={2024},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/Christim/ellesia-gpt-oss-120b}}
}

"Sanctify them through thy truth: thy word is truth." - John 17:17

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. 🙋 Ask for provider support

Model tree for Christim/ellesia-gpt-oss-120b

Finetuned
(73)
this model

Dataset used to train Christim/ellesia-gpt-oss-120b