MRockatansky/Gemma-4-31B-Storymaxxed3
MRockatansky/Gemma-4-31B-Storymaxxed3 is a 31 billion parameter language model fine-tuned from llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic, featuring a 32K context length. Optimized using Direct Preference Optimization (DPO) on a high-quality narrative dataset, this model excels at creative writing, narrative prose, and dialogue generation. It is specifically designed to produce detailed stories with improved prose and scene tracking for both SFW and NSFW scenarios.
Loading preview...
Model Overview
MRockatansky/Gemma-4-31B-Storymaxxed3 is a 31 billion parameter language model, fine-tuned from the llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic base model. It leverages a 32,768 token context length and is specifically optimized for creative writing and narrative generation.
Key Capabilities
- Enhanced Narrative Prose: Produces high-quality, detailed stories with improved prose compared to previous Storymaxxed iterations.
- Superior Dialogue Generation: Excels at creating natural and engaging dialogue within narratives.
- Scene and Detail Tracking: Demonstrates strong ability to maintain consistent scene details and overall narrative coherence.
- Versatile Content Generation: Capable of generating both SFW and NSFW scenarios effectively.
Training Details
The model was trained using TRL with Direct Preference Optimization (DPO) on a high-quality dataset of narrative preference pairs. This method, detailed in the paper "Direct Preference Optimization: Your Language Model is Secretly a Reward Model" (paper link), focuses on aligning the model's output with human preferences for narrative quality.
Recommended Settings
For optimal inference, the following generation parameters are suggested:
- Temperature: 1.0
- Top P: 0.95
- Top K: 64
Use Cases
This model is ideal for applications requiring advanced creative writing, story generation, role-playing, and any task where high-quality narrative prose and dialogue are critical.