Drezzman/Qwen3-4B-Instruct-2507-heretic-uncens
Drezzman/Qwen3-4B-Instruct-2507-heretic-uncens is a 4 billion parameter instruction-tuned causal language model, based on the Qwen3-4B-Instruct-2507 architecture by Qwen, with a native context length of 262,144 tokens. This specific version has been decensored using the Heretic tool, resulting in significantly reduced refusals compared to the original model. It is optimized for general capabilities including instruction following, logical reasoning, mathematics, coding, and long-context understanding, making it suitable for applications requiring less restrictive content generation.
Loading preview...
Drezzman/Qwen3-4B-Instruct-2507-heretic-uncens: Decensored Qwen3-4B-Instruct
This model is a 4 billion parameter instruction-tuned causal language model, derived from the Qwen3-4B-Instruct-2507 developed by Qwen. Its primary differentiator is that it has been decensored using the Heretic v1.2.0 tool, significantly reducing content refusals from 100/100 in the original model to 23/100 in this version.
Key Capabilities & Enhancements
- Decensored Output: Engineered to provide less restrictive content generation, with a notable reduction in refusal rates.
- Enhanced General Capabilities: Features significant improvements in instruction following, logical reasoning, text comprehension, mathematics, science, coding, and tool usage.
- Long-Context Understanding: Supports a native context length of 262,144 tokens, enabling advanced long-form text processing.
- Multilingual Knowledge: Offers substantial gains in long-tail knowledge coverage across multiple languages.
- User Alignment: Demonstrates markedly better alignment with user preferences in subjective and open-ended tasks, leading to more helpful and higher-quality text generation.
Performance Highlights
Compared to the original Qwen3-4B-Instruct-2507, this model maintains the strong performance of its base, which shows leading results across various benchmarks including MMLU-Pro (69.6), GPQA (62.0), AIME25 (47.4), and Creative Writing v3 (83.5). The decensoring process specifically targets output restrictions rather than core performance metrics.
Good For
- Applications requiring less constrained or "uncensored" text generation.
- Tasks demanding strong instruction following and logical reasoning.
- Scenarios benefiting from extensive long-context understanding.
- Multilingual content generation and comprehension.