ConicCat/Gemma4-GarnetV2-31B
ConicCat/Gemma4-GarnetV2-31B is a 31 billion parameter Gemma 4-based language model with a 32768 token context length, primarily fine-tuned by ConicCat to enhance prose and writing capabilities. It excels in creative domains like roleplay and longform text generation, significantly reducing 'AI slop' and improving output diversity. This model is optimized for generating extended, high-quality written content with fewer refusals.
Loading preview...
ConicCat/Gemma4-GarnetV2-31B: Enhanced Creative Writing Model
This model is a 31 billion parameter fine-tune of the Gemma 4 architecture, developed by ConicCat, with a primary focus on improving prose and general writing capabilities. It is particularly well-suited for creative applications, including roleplay and generating diverse, longform text.
Key Capabilities
- Improved Longform Writing: Capable of generating up to 4000 words of text in a single output, making it ideal for extended content creation.
- Reduced 'AI Slop': Markedly decreases generic 'AI-isms' and improves the naturalness of generated text.
- Enhanced Output Diversity: Offers a wider range of stylistic and content variations in its responses.
- Fewer Refusals: Exhibits fewer 'soft' refusals when generating creative or imaginative content.
Training & Differentiation
The V2 iteration of Garnet incorporates more and better roleplay data, alongside a shift towards using fantasy and sci-fi books for training over literary fiction. While not its primary focus, the model includes instruct data from internlm/Condor-SFT-20K to maintain general intellect and instruction-following abilities. It also utilizes a reformatted version of ConicCat/Gutenberg-SFT and proprietary backtranslated book datasets.
Good For
- Creative Writing: Generating stories, scripts, and other imaginative content.
- Roleplay Scenarios: Providing detailed and diverse responses in interactive roleplaying.
- Longform Content Generation: Producing extended articles, narratives, or descriptive passages.