ewald1976/gemma-4-12b-crownelius-writer

Gemma-4-12b-it-crownelius-writer

This model is a Supervised Fine-Tuning (SFT) variant of Gemma-4-12b-it, trained for 3 epochs using Unsloth on the custom Crownelius datasets.

The primary goal of this fine-tuning was to shift the model's prose away from typical over-energetic, repetitive AI tropes toward a more mature, nuanced, and grounded creative writing style.

Model Description

• Base Model: unsloth/gemma-4-12b-it
• Training Method: SFT (Supervised Fine-Tuning) via Unsloth
• Epochs: 3
• Primary Focus: Creative Writing, Novel Crafting, Style Transformation

Style Shift

Standard instructional models often tend to write with an overly enthusiastic, "caffeinated" tone. By introducing the Crownelius datasets, this fine-tune refines the output to be more patient, structurally sound, and sophisticated.

Training Data

The model was trained on a curated mix of high-quality creative writing samples:

• Crownelius/Creative-Writing-High-Quality-1300x
• Crownelius/Novel-Crafting-250x

Usage & Prompting

Since this is an SFT of an instruct-tuned base, it adheres to the standard Gemma-4 chat template. For optimal creative results, try using system prompts that encourage descriptive pacing.