DavidAU/Qwen3-8B-64k-Context-2X-Josiefied-Uncensored

Overview

This model, DavidAU/Qwen3-8B-64k-Context-2X-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", specifically enhanced to support a significantly larger context window of 64k (65536) tokens through the application of YARN techniques. The model is provided in "safe tensors" format, enabling conversion to various quantization formats like GGUF, GPTQ, EXL2, AWQ, and HQQ.

Key Capabilities & Features

• Extended Context Window: Supports 64k tokens, a substantial increase from the original 32k, allowing for processing of much longer inputs and maintaining coherence over extended conversations or documents.
• Reasoning Optimization: Designed for deep thinking and systematic reasoning, with an optional system role prompt to encourage detailed internal monologues and structured problem-solving.
• Uncensored: Based on an "abliterated" version, suggesting a less restrictive content policy.
• Quantization Support: Source code is provided for generating various quantized formats, including GGUF, for broader compatibility and efficient deployment across different hardware.

Usage Recommendations

• Optimal Performance: Users are strongly advised to consult the linked "Maximizing-Model-Performance-All..." guide for critical parameter, sampler, and advanced sampler settings to achieve the best results.
• System Role: An optional system role is provided to activate deep thinking and chain-of-thought reasoning, enclosed within <think> tags.
• Templating: Recommends using Jinja or CHATML templates for interaction.
• Minimum Context: Suggests a minimum context limit of 8k to 16k for optimal "thinking" and output generation.