Gryphe/Gemma-4-12B-StyleTune
Gryphe/Gemma-4-12B-StyleTune is a 12 billion parameter language model from the Gemma 4 family, developed by Gryphe. This model features a unique "style tune" approach, where only the lm_head output projection layer is trained, resulting in a 56% reduction in clichés and a distinct writing style while preserving all original Gemma capabilities. With a 32768 token context length, it is optimized for narrative generation and creative writing tasks requiring a specific voice.
Loading preview...
Gryphe/Gemma-4-12B-StyleTune: A Unique Approach to Stylistic Control
Gryphe/Gemma-4-12B-StyleTune is a 12 billion parameter model from the Gemma 4 family, distinguished by its innovative "style tune" methodology. Unlike traditional fine-tuning that modifies most model parameters, this model exclusively trains the lm_head output projection layer. This targeted approach significantly alters the model's writing style and vocabulary without impacting its core reasoning, world knowledge, or instruction-following capabilities.
Key Capabilities & Differentiators
- Stylistic Transformation: Achieves a 56% reduction in clichés and an almost entirely different trigram vocabulary (only 16.8% shared with the base model), offering a fresh narrative voice.
- Preserved Core Intelligence: All of Gemma's original reasoning, language understanding, and instruction-following abilities remain fully intact, as only the stylistic output layer is modified.
- Efficient Fine-tuning: The method requires significantly less VRAM and training time, making it accessible for consumer hardware while delivering substantial stylistic changes.
- Narrative Focus: Trained on 100% narrative data, specifically designed to enhance creative writing and roleplay scenarios.
Ideal Use Cases
This model is particularly well-suited for applications where a distinct and refined writing style is paramount, such as:
- Creative Writing: Generating stories, prose, or dialogue with a unique and less cliché-ridden voice.
- Roleplay Scenarios: Producing engaging and stylistically consistent character responses.
- Content Generation: Creating narrative content that requires a specific tone or avoids common linguistic patterns.