nurdich/Qwen3.8-9B-Distill-SLERP-F451-Pro-Writer-Uncensored
nurdich/Qwen3.8-9B-Distill-SLERP-F451-Pro-Writer-Uncensored is a 9 billion parameter Qwen3.5-based model merged using the SLERP method, specifically optimized for creative writing, role-playing, and Chinese NSFW content. It combines a Qwen3.8 distillation focused on general intelligence (math, code, reasoning) with a Qwen3.5 model specialized in creative prose style. This model leverages a 32768 token context length to enhance its writing capabilities.
Loading preview...
Model Overview
This model, created by nurdich, is a 9 billion parameter merge of two Qwen3.5-based models using the SLERP method. It combines the strengths of a general intelligence model with a specialized creative writing model, aiming to provide robust performance in both areas. The merge was performed using Mergekit and Kaggle's free sessions.
Key Capabilities
- Creative Writing: Specifically fine-tuned for generating creative prose and enhancing writing style.
- Role-playing: Optimized for engaging in role-playing scenarios.
- Chinese NSFW Content: Noted for its performance in Chinese NSFW contexts.
- General Intelligence: Incorporates a Qwen3.8 distillation focused on reasoning, mathematical tasks, code generation, and tool use.
- Uncensored Output: The parent models are uncensored/ablated, meaning outputs are not safety-filtered.
Merge Details
The model integrates petruhonk/Qwen3.8-9B-Distill-uncensored-heretic (general intelligence, math, code, reasoning) as the base and DavidAU/Qwen3.5-9B-The-Bradbury-F451-Pro-Writer-Uncensored-Heretic (creative writing specialist). A layer-wise t schedule was applied during the SLERP merge, with early layers leaning towards the distilled reasoning model and later attention layers favoring the writer model, aiming to preserve general intelligence while adding creative writing prowess. The model uses bfloat16 dtype.