Hugging Face Overview
Overview
DavidAU/Qwen3-4B-Fiction-On-Fire-Series-7-Model-X is a 4 billion parameter model built on the Qwen3 architecture, created by DavidAU. It is a high-precision merge that has undergone random pruning to specifically alter and improve its prose, creativity, and general knowledge. This model is part of the 'Fiction on Fire' Series 7, where each model is unique due to varying random pruning densities.
Key Capabilities
- Enhanced Creative Writing: Specifically tuned for generating vivid and graphic prose, as demonstrated by its performance in horror scene generation.
- Decensored Base: Built upon "Josiefied-Qwen3-4B-abliterated-v2" to minimize refusals, though refusal rates may vary between models in the series due to pruning.
- Robust Reasoning: Maintains full and functional reasoning capabilities across all models in the series.
- Extended Context: Supports a context length of 40,000 tokens, with the potential for further alteration using 'yarn' for creative use cases.
- Customizable Output: Performance and output characteristics can be significantly influenced by temperature, repetition penalty, and other sampling parameters, with specific recommendations provided for creative and non-creative tasks.
Good For
- Creative Writing & Roleplay: Excels in generating detailed and immersive narrative content, particularly for genres like horror.
- Scene Generation: Optimized for creating and continuing scenes with rich descriptive language and emotional impact.
- Experimentation with Merges: Users can leverage this model for further merging experiments, as its unique pruning methodology offers diverse starting points.
- Long-form Content: Its substantial context window makes it suitable for generating extended narratives and complex story arcs.