GemmaBible

A fine-tuned Gemma 4 E4B model specialized in biblical scholarship, theology, and Bible study. Grounded in the Berean Standard Bible (BSB).

What it does

• Quotes Scripture precisely from the BSB with proper citations
• Provides Greek and Hebrew word studies with transliterations and Strong's numbers
• Presents Protestant, Catholic, and Orthodox perspectives on debated topics
• Analyzes passages with scholarly hermeneutics and cited sources
• Detects and corrects common misquotations
• Stays within theological boundaries — declines non-theological requests

Training

• Base model: google/gemma-4-E4B-it (4.5B effective parameters)
• Method: QLoRA (rank 64, alpha 64) with Unsloth
• Data: ~8,000 instruction examples across 11 specialized generators covering comparative theology, Greek/Hebrew exegesis, systematic theology, creedal analysis, and more
• Hardware: NVIDIA RTX PRO 6000 (96GB)
• Epochs: 3 | Final loss: 0.40

Usage

With Ollama (GGUF)

Download merged.Q5_K_M.gguf and create a Modelfile:

FROM ./merged.Q5_K_M.gguf
PARAMETER temperature 0.3
PARAMETER top_p 0.9
PARAMETER num_ctx 4096

ollama create gemmabible -f Modelfile
ollama run gemmabible "What does John 3:16 mean?"

With Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("rhemabible/GemmaBible")
tokenizer = AutoTokenizer.from_pretrained("rhemabible/GemmaBible")

Files

File	Format	Size	Use
model.safetensors	SafeTensors	~15 GB	Full precision weights
merged.Q5_K_M.gguf	GGUF	~5.5 GB	Ollama / LM Studio / llama.cpp

Limitations

• Trained on BSB text; may be less accurate with other Bible translations
• 4.5B parameters — less capacity for nuanced multi-turn theological debate compared to larger models
• Not a substitute for pastoral counsel or formal theological education