DavidAU/Qwen3-8B-192k-Context-6X-Josiefied-Uncensored
DavidAU/Qwen3-8B-192k-Context-6X-Josiefied-Uncensored is an 8 billion parameter Qwen3-based causal language model, derived from Goekdeniz-Guelmez's "Josiefied-Qwen3-8B-abliterated-v1." This model significantly extends the original 32k context to 192k tokens using YARN, making it suitable for tasks requiring extensive context processing. It is designed for long-form output and creative generation, with specific recommendations for optimal inference settings.
Loading preview...
Model Overview
This model, DavidAU/Qwen3-8B-192k-Context-6X-Josiefied-Uncensored, is an 8 billion parameter variant of the Qwen3 architecture. It is a modification of Goekdeniz-Guelmez's "Josiefied-Qwen3-8B-abliterated-v1" and has been specifically enhanced to support an extended context window.
Key Differentiators
- Extended Context Length: The primary feature is its significantly increased context window, expanded from the base 32k (32768) tokens to 192k (196608) tokens using the YARN technique, as detailed in the Qwen repository's technical notes. This allows for processing and generating much longer sequences of text.
- Josiefied Base: Built upon the "Josiefied-Qwen3-8B-abliterated-v1" fine-tune, suggesting a focus on specific performance characteristics or instruction following.
- Uncensored: The "Uncensored" tag indicates a model designed with fewer content restrictions, potentially offering broader utility for various applications.
Recommended Usage
- Optimal Settings: For best performance, especially in creative and long-form output, specific inference parameters are recommended: Temperature 1+, Top K 100+, and Repetition Penalty 1.02-1.09.
- Templating: Users should employ either Jinja or CHATML templates for interaction.
- System Role: An optional system role is provided for deep thinking and systematic reasoning, allowing the model to deliberate internally before providing a solution, enclosed in
<think> </think>tags.
Performance Optimization
DavidAU emphasizes reviewing the "Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters" document for critical parameter, sampler, and advanced sampler settings to enhance operation across various AI/LLM applications, regardless of the model's class or quantization format.