DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP
DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP is a multi-stage fine-tuned and merged Qwen 3.6 27B model, developed by DavidAU in collaboration with Nightmedia, TeichAI, armand0e, and trohrbaugh. This model is notable for being the first of its size to exceed 700 on the ARC-C benchmark in both 8-bit and 4-bit quantization, outperforming base Qwen 3.6 27B and Qwen3.6-35B-A3B across most benchmarks. It is designed to significantly enhance general intelligence, instruction following, and problem-solving abilities, making it suitable for a wide range of demanding analytical and reasoning tasks.
Loading preview...
Model Overview
DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP is a highly optimized, multi-stage fine-tuned and merged model based on the Qwen 3.6 27B architecture. Developed through a collaborative effort by DavidAU, Nightmedia, TeichAI, armand0e, and trohrbaugh, this model aims to push the boundaries of open-source LLM performance on consumer hardware.
Key Differentiators
- Intelligence Benchmark Leader: This model is the first of its size to achieve an ARC-C score exceeding 700 in both 8-bit and 4-bit quantization, a performance level previously associated with closed-source models like OpenAI, Claude, and Gemini.
- Superior Performance: It surpasses the base Qwen 3.6 27B in 6 out of 7 benchmarks and matches it on the seventh, while exceeding all 7 benchmarks for Qwen3.6-35B-A3B.
- Enhanced Core Capabilities: The primary focus during its creation was to significantly improve general intelligence, instruction following, and problem-solving abilities without compromising the core model integrity.
- Multi-Stage Training: Incorporates light traces from 'Fable', Claude Opus (for reasoning), F451 (in-house dataset), and GPT5 (Polaris, non-reasoning) through a complex multi-stage fine-tuning and merging process.
- Uncensored: Features 'Heretic' uncensoring, providing a less restrictive output while maintaining the model's core capabilities.
- Vision Capable: Supports vision inputs, requiring a separate 'mmproj' file for activation.
- Optimized Quantizations: Offers NEO IMATRIX GGUF quants (regular and MTP) with improved accuracy and long context performance, including specialized 'LOW' and 'AMD/VULCAN' versions.
Ideal Use Cases
- Complex Problem Solving: Excels in tasks requiring advanced reasoning and analytical capabilities.
- Instruction Following: Highly effective for applications demanding precise adherence to instructions.
- General AI Applications: Suitable as a robust, all-purpose model for various demanding tasks where high intelligence is paramount.
- Consumer Hardware Deployment: Optimized for strong performance on consumer-grade hardware, making it accessible for a wider range of developers.