avi686/Qwen3-14B-heretic

Qwen3-14B-heretic: Decensored and Enhanced

This model, avi686/Qwen3-14B-heretic, is a 14.8 billion parameter causal language model derived from Qwen/Qwen3-14B, specifically modified using the Heretic v1.3.0 tool to be a decensored version. It maintains the robust architecture of the Qwen3 series while significantly reducing content refusals, demonstrating 5 refusals out of 100 compared to 99/100 in the original model.

Key Capabilities

• Dual-Mode Operation: Uniquely supports seamless switching between a 'thinking mode' for complex logical reasoning, mathematics, and code generation, and a 'non-thinking mode' for efficient, general-purpose dialogue. This can be controlled via enable_thinking parameter or /think and /no_think tags in prompts.
• Enhanced Reasoning: Shows significant improvements in mathematical problem-solving, code generation, and commonsense logical reasoning, surpassing previous Qwen models.
• Superior Human Preference Alignment: Excels in creative writing, role-playing, multi-turn dialogues, and instruction following, offering a more natural conversational experience.
• Advanced Agentic Functions: Demonstrates strong capabilities for tool calling and integration with external tools, achieving leading performance among open-source models in complex agent-based tasks.
• Multilingual Support: Supports over 100 languages and dialects with robust multilingual instruction following and translation abilities.
• Extended Context Length: Natively handles up to 32,768 tokens, with validated performance up to 131,072 tokens using the YaRN method for long text processing.

Good for

• Applications requiring reduced content moderation or censorship compared to the base Qwen3-14B model.
• Tasks demanding complex logical reasoning, mathematical problem-solving, or code generation where the 'thinking mode' can be leveraged.
• Creative writing, role-playing, and multi-turn conversational agents that benefit from superior human preference alignment.
• Multilingual applications including instruction following and translation across a wide array of languages.
• Agent-based systems that require precise tool integration and high performance in complex tasks.
• Scenarios requiring long context processing, especially with the YaRN extension for up to 131,072 tokens.