Model Overview
DavidAU/Qwen3-Short-Story-Instruct-Uncensored-262K-ctx-4B is a 4 billion parameter instruction-tuned model, fine-tuned by DavidAU from a Josiefied-Qwen3-4B-Instruct base. This model has undergone a specialized fine-tuning process (15% of the model for 5 epochs) using a short story dataset, specifically enhancing its prose generation and general creative writing abilities. It is designed to produce unique and detailed short stories, with distinct word choices and increased narrative depth.
Key Capabilities
- Short Story Generation: Optimized to output short stories with a unique style and limited output size.
- Enhanced Prose & Creativity: Alters prose generation and creative abilities, offering increased depth and detail in narratives.
- Uncensored Content Generation: While not explicitly uncensored by default, it can be directed to generate graphic, cursing, or explicit content with specific prompting.
- High Context Length: Supports a maximum context of 256k (as stated in the README, though the model name indicates 40960 tokens).
Recommended Usage
- Creative Writing: Ideal for generating fictional short stories, scenes, and creative prose.
- Exploratory Prompting: Suggested temperatures of 0.8 to 1.5 and a repetition penalty of 1.05 to 1.1 are recommended for varied outputs.
- Attitude & Refusals: Users should be aware the model may have an "attitude" and might require regeneration for desired length or to bypass initial refusals.
- Jinja/CHATML Template: Requires a Jinja (embedded) or CHATML template for optimal performance.