laurentiurad/Qwen3-4b-decensored-instruct is a 4 billion parameter causal language model, derived from Qwen/Qwen3-4B-Instruct-2507 and decensored using the Heretic tool. This model features significant improvements in general capabilities, including instruction following, logical reasoning, and long-tail knowledge coverage across multiple languages, while offering a native context length of 262,144 tokens. Its primary differentiator is its "decensored" nature, achieved by modifying the original Qwen3-4B-Instruct-2507 model to reduce refusals.
Loading preview...
Model Overview
laurentiurad/Qwen3-4b-decensored-instruct is a 4 billion parameter causal language model, built upon the Qwen3-4B-Instruct-2507 base model. This version has been specifically processed using the Heretic v1.2.0 tool to achieve a "decensored" state, significantly reducing model refusals from 93/100 in the original to 5/100 in this variant. It maintains the original model's robust capabilities and impressive context length.
Key Capabilities & Enhancements
- Decensored Output: The primary feature is its reduced tendency to refuse prompts, making it more permissive than its base model.
- General Capability Improvements: Inherits significant enhancements in instruction following, logical reasoning, text comprehension, mathematics, science, coding, and tool usage from the Qwen3-4B-Instruct-2507 base.
- Extended Context Length: Supports a native context length of 262,144 tokens, enabling processing of very long inputs.
- Multilingual Support: Offers substantial gains in long-tail knowledge coverage across multiple languages.
- User Alignment: Provides markedly better alignment with user preferences in subjective and open-ended tasks, leading to more helpful responses.
Performance Highlights
Compared to the original Qwen3-4B-Instruct-2507, this model demonstrates a KL divergence of 1.2951, indicating the changes introduced by the decensoring process. The base model itself shows strong performance across various benchmarks, including MMLU-Pro (69.6), GPQA (62.0), AIME25 (47.4), and Creative Writing v3 (83.5), often outperforming other models in its class. This decensored version aims to retain these performance characteristics while offering greater flexibility in content generation.
Use Cases
This model is suitable for applications requiring a powerful 4B parameter LLM with an exceptionally long context window and a reduced propensity for content refusals. It excels in complex instruction following, detailed reasoning, and generating creative or open-ended text across multiple languages.