MuXodious/Qwen2.5-7B-Instruct-1M-Thinking-Claude-Gemini-GPT5.2-DISTILL-PaperWitch-heresy
MuXodious/Qwen2.5-7B-Instruct-1M-Thinking-Claude-Gemini-GPT5.2-DISTILL-PaperWitch-heresy is a 7.6 billion parameter instruction-tuned Qwen2.5-7B-Instruct-1M model, fine-tuned by MuXodious using P-E-W's Heretic engine. It integrates high-reasoning datasets from Claude Opus 4.5, Gemini, and GPT5.2, along with a 1 million context window, to produce compact and high-quality reasoning blocks. This model is specifically optimized for enhanced thinking and reasoning capabilities, directly improving output generation in terms of detail, length, complexity, and overall quality, making it suitable for tasks requiring deep analytical processing.
Loading preview...
Overview
MuXodious/Qwen2.5-7B-Instruct-1M-Thinking-Claude-Gemini-GPT5.2-DISTILL-PaperWitch-heresy is a 7.6 billion parameter instruction-tuned model based on Qwen2.5-7B-Instruct-1M. It has been fine-tuned using P-E-W's Heretic (v1.2.0) ablation engine with Magnitude-Preserving Orthogonal Ablation. This model incorporates three high-reasoning fine-tuning datasets from Claude Opus 4.5, Gemini, and GPT5.2, along with a 1 million context window, to convert the base model into a "thinking/reasoning" model.
Key Capabilities
- Enhanced Reasoning: Produces compact and "to the point" reasoning blocks, which significantly improve the quality of final output generation.
- High-Quality Output: Directly improves output generation in terms of detail, length, complexity, and overall quality.
- Temperature Robustness: Thinking/Reasoning activation is not significantly affected by temperature settings, allowing for a wide range (0.1 to 2.5 or higher).
- Self-Generating Thinking Tags: Does not require a system prompt; thinking tags/blocks will self-generate.
- Optimized for Specific Quants: Suggested quants are Q4KS (non-imatrix) or IQ3_M (imatrix) or higher for optimal reasoning activation.
Good For
- Complex Problem Solving: Excels in tasks requiring deep analytical processing and structured reasoning.
- Creative Content Generation: Recommended for creative tasks with a minimum temperature of 1.2 or higher.
- Detailed Output Generation: Ideal for scenarios where detailed, complex, and high-quality responses are crucial.
- Applications Requiring Consistent Reasoning: Suitable for use cases where consistent and compact reasoning is preferred over verbose explanations.
Note: For optimal performance, a minimum temperature of 0.7 (1.2+ for creative tasks) and a repetition penalty of 1.05 (adjustable to 1 for longer reasoning blocks) are suggested. A minimum context window of 4k, preferably 8k+, is also recommended. If thinking does not activate, users can regenerate or prepend prompts with "Think Deeply: ".