Overview

This model, maldocray/Qwen3-4B-Instruct-2507-heretic, is a 4.0 billion parameter instruction-tuned causal language model derived from the Qwen3 architecture by Qwen. It features a substantial native context length of 262,144 tokens, though a context length of 32,768 is recommended to avoid out-of-memory issues. A key differentiator of this specific model is its decensored nature, achieved using Heretic v1.0.0, which significantly reduces its refusal rate (21/100) compared to the original Qwen3-4B-Instruct-2507 (99/100).

Key Capabilities

• Enhanced General Capabilities: Demonstrates significant improvements in instruction following, logical reasoning, text comprehension, mathematics, science, coding, and tool usage.
• Long-Context Understanding: Supports 256K long-context understanding, enabling processing of extensive inputs.
• Improved Alignment: Offers markedly better alignment with user preferences for subjective and open-ended tasks, leading to more helpful and higher-quality text generation.
• Multilingual Support: Provides substantial gains in long-tail knowledge coverage across multiple languages.
• Agentic Use: Excels in tool-calling capabilities, with recommended integration via Qwen-Agent.

Good For

• Applications requiring a model with reduced content moderation and lower refusal rates.
• Tasks demanding strong performance in logical reasoning, mathematical problem-solving, and scientific inquiry.
• Code generation and understanding, as well as tool-use scenarios.
• Generating high-quality, helpful, and open-ended text responses across various subjects and languages.
• Use cases benefiting from a very long context window for complex or extensive inputs.