0xSojalSec/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU
The 0xSojalSec/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU model is a 27 billion parameter language model developed by DavidAU, based on the Qwen 3.8 architecture. It is fine-tuned for enhanced reasoning, reduced 'thinking tokens', and uncensored output, while maintaining high detail and stability. This model excels in complex analytical tasks and creative generation, demonstrating significant performance improvements over base Qwen 3.8 27B benchmarks, particularly in ARC-C scores.
Loading preview...
Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU
This model is the first release in a series of highly optimized 27 billion parameter language models by DavidAU, built upon the Qwen 3.8 architecture. It features extensive fine-tuning across multiple stages, focusing on enhancing core metrics, reducing computational overhead, and providing uncensored output while preserving model integrity.
Key Capabilities
- Superior Benchmarking: Achieves significantly higher scores than base Qwen 3.8 27B, with ARC-C scores 141 points above the baseline. Performance in 4-bit quantization is nearly identical to 8-bit.
- Optimized Reasoning: Demonstrates a strong reduction in 'thinking tokens' (1/2 to 1/10 of normal Qwen size) while maintaining or increasing detail level, with auto-variable thinking sizes based on prompt complexity.
- Uncensored Output: Undergoes a 'heretic'ing' process to remove safety alignments, balanced with a low KLD (Kullback-Leibler Divergence) to minimize performance degradation, followed by a 'healing' training step to restore and enhance core metrics.
- Stability: Stress-tested in 4-bit, non-imatrix configurations to ensure robust and reliable performance.
- Enhanced Character: Exhibits changes in output character, specifically in depth of thinking and analytical capabilities, leading to more detailed and nuanced responses.
Good For
- Complex Analytical Tasks: Ideal for applications requiring deep reasoning and detailed analysis, where traditional models might overthink or provide less concise outputs.
- Creative Content Generation: Excels in generating highly detailed and creative narratives, as demonstrated by its ability to produce engaging and vivid prose.
- Unrestricted Content Creation: Suitable for use cases where uncensored and unfiltered responses are desired, without compromising factual accuracy or detail.
- Resource-Efficient Deployment: Its strong 4-bit performance makes it a viable option for environments with computational constraints, offering near 8-bit quality at reduced resource cost.