Ateron/Gemma-4-Novelist-Eclipse-31B

VISIONConcurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 5, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Ateron/Gemma-4-Novelist-Eclipse-31B is a 31 billion parameter language model based on the Gemma-4 architecture, developed by Ateron. This model is a merge of several fine-tuned models, specifically engineered for creative writing tasks with a focus on infusing creativity and stability into prose generation. It aims to produce detailed descriptions and consistent narrative output, making it suitable for applications requiring advanced text generation capabilities.

Loading preview...

Overview of Ateron/Gemma-4-Novelist-Eclipse-31B

Ateron's Novelist-Eclipse is a 31 billion parameter language model built upon the Gemma-4 architecture, designed to excel in creative writing. This model is the result of a multi-phase merge process, combining several specialized models to enhance its narrative capabilities and stability.

Key Capabilities

  • Enhanced Creative Writing: Specifically developed to generate imaginative and detailed prose, building on the "Novelist" concept.
  • Improved Stability: The merging process focused on infusing creativity while also ensuring greater consistency in output compared to earlier experimental versions.
  • Description Focus: The initial "dare_ties" merge phase was re-engineered to prioritize descriptive quality, aiming for less "slop" in the generated text.
  • Consistency through Merging: A second "model_stock" phase integrated additional models to further improve the consistency of the generated content.

Training and Development

The model was synthesized using mergekit, involving two distinct phases:

  1. Phase 1: The Prose (dare_ties): This phase focused on description, merging models like G4-Gutenberg, Melinoe, and Glimmer with specific density and weight parameters.
  2. Phase 2: The Lead (model_stock): This phase aimed to improve consistency by merging NovelistX, GarnetV2, and Gemopus. The base model for both phases was Gemma-4-it.

Good For

  • Creative writing applications requiring detailed and imaginative text generation.
  • Generating consistent narrative content.
  • Tasks where descriptive quality is paramount.

Finetunes
1 models