Gryphe/Gemma-4-26B-A4B-StyleTune-V2

VISIONConcurrency Cost:2Model Size:26BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 20, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Gryphe/Gemma-4-26B-A4B-StyleTune-V2 is a 26 billion parameter Gemma 4-based language model with a 32768 token context length, specifically fine-tuned by Gryphe to modify its writing style. This model achieves a 52% reduction in clichés and a significantly different trigram vocabulary compared to the base instruct model, while retaining all original reasoning and instruction-following capabilities. It is optimized for narrative generation and creative writing applications where a distinct and less cliché-ridden voice is desired.

Loading preview...

Gryphe/Gemma-4-26B-A4B-StyleTune-V2: Targeted Style Tuning

Gryphe/Gemma-4-26B-A4B-StyleTune-V2 is a 26 billion parameter model built upon the Gemma 4 architecture, featuring a 32768 token context length. This model represents a unique "style tune" approach, where only the lm_head output projection tensor is trained, leaving all other transformer layers and capabilities of the base Gemma 4 model completely intact. This method significantly reduces VRAM requirements and training time, making it accessible on consumer hardware.

Key Capabilities & Differentiators

  • Targeted Style Modification: Achieves a 52% reduction in clichés and a 19.9% shared trigram vocabulary compared to the base instruct model, resulting in a distinct and less repetitive writing style.
  • Preserved Core Intelligence: All original reasoning, world knowledge, instruction following, and language understanding capabilities of the Gemma 4 26B-A4B model are fully retained.
  • Efficient Fine-tuning: The single-tensor training approach allows for rapid fine-tuning without altering the model's fundamental intelligence.
  • Narrative Optimization: Trained exclusively on narrative data, making it highly suitable for creative writing and roleplay scenarios.

Ideal Use Cases

  • Creative Writing: Generating unique and engaging narratives, stories, and descriptive text.
  • Roleplay: Producing responses with a distinct and less generic voice.
  • Content Generation: Creating varied and original content where stylistic nuance is important.

This model is designed for users who appreciate the robust capabilities of Gemma 4 but desire a more refined and less cliché-prone output style, without compromising on the underlying intelligence.