Endersand/Gemma-4-12B-StyleTune

TEXT GENERATIONPricing:Input $1.2 / Cached $0.24 / Output $4.8Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Endersand/Gemma-4-12B-StyleTune is a 12 billion parameter language model from Endersand, part of the Gemma 4 family, specifically optimized for narrative writing style. This model achieves a 56% reduction in clichés and a distinct writing voice by exclusively training the lm_head output projection. It retains all of Gemma's original reasoning and instruction-following capabilities, making it ideal for creative text generation where style is paramount.

Loading preview...

Endersand/Gemma-4-12B-StyleTune Overview

Endersand/Gemma-4-12B-StyleTune is a 12 billion parameter model from the Gemma 4 family, uniquely fine-tuned to alter writing style without impacting core capabilities. Unlike traditional finetuning, this model focuses exclusively on training the lm_head output projection layer, which dictates token emission and, consequently, the model's voice. This targeted approach significantly reduces VRAM requirements and training time, allowing for efficient style modification on consumer hardware.

Key Capabilities & Differentiators

  • Distinct Writing Style: Achieves a 56% reduction in clichés and an almost entirely different trigram vocabulary (only 16.8% shared with the base model), resulting in a less generic and more unique narrative output.
  • Preserved Core Intelligence: All of Gemma 4's original reasoning, world knowledge, instruction following, and language understanding capabilities remain fully intact, as only the style-determining lm_head was modified.
  • Efficient Tuning: The 'style tune' method allows for significant stylistic changes with minimal computational overhead, making it accessible for developers with limited resources.

Ideal Use Cases

  • Creative Writing: Generating narrative content, stories, or roleplay scenarios where a unique and cliché-free writing style is desired.
  • Content Generation: Producing text with a specific voice or tone without sacrificing the underlying factual or logical coherence of the base Gemma model.
  • Experimentation: Exploring the impact of targeted finetuning on specific model components for stylistic control.

Finetunes
1 models