Ateron/Gemma-4-AssGuard-31B

VISIONConcurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Ateron/Gemma-4-AssGuard-31B is a 31 billion parameter language model based on the Gemma-4 architecture, developed by Ateron through a multi-phase mergekit process. This model is specifically designed to generate humorous and engaging outcomes, making it suitable for creative text generation and experimental applications. It leverages a unique merging strategy to combine various fine-tuned models, aiming for distinctive and entertaining outputs.

Loading preview...

Overview

Ateron/Gemma-4-AssGuard-31B is a 31 billion parameter model created by Ateron, primarily for experimental and fun text generation. It is built upon the Gemma-4 architecture and developed using a multi-phase mergekit process, combining several base models to achieve its unique output style. The creator notes that the model can produce "very funny outcomes" and aims to "steal your attention for several swipes."

Key Characteristics

  • Gemma-4 Architecture: Based on Google DeepMind's Gemma-4 base model.
  • Mergekit Synthesis: Developed through a three-phase mergekit process, including dare_ties and model_stock merge methods, integrating models like G4-Gutenberg, Pantheon-Reasoning-31B-1.1, Equinox, Gemma-4-31B-storymaxxed2, and GarnetV2.
  • Humorous Output: Designed to generate engaging and often humorous text.
  • Recommended Settings: The creator suggests using a high min_p value, specifically between 0.15-0.20, for optimal results.

Use Cases

  • Creative Text Generation: Ideal for generating entertaining and funny content.
  • Experimental Applications: Suitable for users looking to explore unique model behaviors and outputs.
  • Interactive Storytelling: Can be used in scenarios where a distinctive and engaging conversational style is desired.

Finetunes
1 models