MRockatansky/Gemma-4-31B-Storymaxxed3
MRockatansky/Gemma-4-31B-Storymaxxed3 is a fine-tuned Gemma-4 model, based on llmfan46's Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic, specifically optimized for creative writing and narrative prose. This model excels at generating high-quality stories, narratives, and dialogue, with improved prose and detailed scene tracking. It was trained using Direct Preference Optimization (DPO) on a dataset of narrative preference pairs, retaining a low refusal rate for creative tasks.
Loading preview...
Model Overview
MRockatansky/Gemma-4-31B-Storymaxxed3 is a specialized fine-tune of the Gemma-4 architecture, building upon the llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic base model. This iteration focuses on enhancing creative writing capabilities, particularly in generating narrative prose and dialogue.
Key Capabilities
- Creative Writing: Optimized for producing quality stories and narratives.
- Improved Prose: Exhibits enhanced word selection and overall writing style compared to previous Storymaxxed versions.
- Dialogue Generation: Capable of generating good, natural-sounding dialogue.
- Scene and Detail Tracking: Excels at maintaining consistent scene details and narrative flow.
- Low Refusal Rate: Retains a low refusal rate (10/100) for creative prompts, similar to its base model.
Training Details
The model was trained using Direct Preference Optimization (DPO) with the TRL library on a high-quality dataset of narrative preference pairs. This method aims to align the model's output with human preferences for narrative quality.
Recommended Usage
For optimal inference results, users should apply the standard generation parameters recommended for Gemma-4 models:
- Temperature: 1.0
- Top P: 0.95
- Top K: 64
A separate mmproj file for vision capabilities is also available for download.