Gryphe/Gemma-4-26B-A4B-StyleTune
Gryphe/Gemma-4-26B-A4B-StyleTune is a 26 billion parameter Gemma 4 A4B model by Gryphe, specifically fine-tuned to alter its writing style. This model focuses on reducing clichés and introducing a new narrative voice by training only the lm_head output projection. It retains all of Gemma 4's original reasoning and instruction-following capabilities, making it suitable for narrative generation and roleplay where a distinct writing style is desired.
Loading preview...
Overview
Gryphe/Gemma-4-26B-A4B-StyleTune is a specialized fine-tune of the Gemma 4 26B-A4B model, developed by Gryphe. Unlike traditional fine-tuning that modifies many layers, this model exclusively trains the lm_head output projection, the final layer responsible for token emission. This unique approach dramatically alters the model's writing style while preserving all of Gemma 4's core capabilities, such as reasoning, world knowledge, and instruction following.
Key Capabilities
- Altered Writing Style: Achieves a 54% reduction in clichés and introduces an almost entirely new trigram vocabulary, resulting in a distinct narrative voice.
- Efficient Fine-tuning: By training only one tensor, it offers significant impact with minimal hardware requirements and faster training times.
- Preserved Core Intelligence: All of Gemma 4's original reasoning, instruction following, and language understanding remain fully intact.
- Narrative Generation: Optimized using 100% narrative data, making it highly suitable for creative writing and roleplay scenarios.
Good For
- Creative Writing: Ideal for generating narrative content with a unique, less clichéd style.
- Roleplay: Excels in scenarios requiring distinct character voices and engaging storytelling.
- Style Experimentation: Developers looking to modify a model's output style without impacting its underlying intelligence.
This version has been superseded by 26B-A4B V2, which is recommended for use.