richardyoung/Qwen2.5-7B-Instruct-heretic

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 24, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The richardyoung/Qwen2.5-7B-Instruct-heretic is a 7.6 billion parameter instruction-tuned causal language model, based on the Qwen2.5 architecture, developed by Richard Young using the Heretic v1.4.0 tool. This model is a decensored version of the original Qwen/Qwen2.5-7B-Instruct, specifically modified to reduce refusals. It supports a context length of up to 32,768 tokens and is designed for general instruction-following tasks with enhanced flexibility.

Loading preview...

Model Overview

This model, richardyoung/Qwen2.5-7B-Instruct-heretic, is a 7.6 billion parameter instruction-tuned causal language model derived from the Qwen2.5 series. It was created by Richard Young using the Heretic v1.4.0 tool, specifically to produce a "decensored" version of the original Qwen/Qwen2.5-7B-Instruct.

Key Differentiators

  • Decensored Output: Significantly reduces refusals compared to the original Qwen2.5-7B-Instruct, with a refusal rate of 3/100 versus 99/100 for the original.
  • Reproducible Modification: The decensoring process is reproducible, with specific abliteration parameters detailed in the README.
  • Enhanced Capabilities (from base Qwen2.5): Inherits improvements from the Qwen2.5 series, including significantly more knowledge, improved coding and mathematics capabilities, better instruction following, and enhanced long text generation (up to 8K tokens output).
  • Long Context Support: Supports a full context length of 131,072 tokens, with the current configuration set for 32,768 tokens, and can generate up to 8,192 tokens. It utilizes YaRN for long context extrapolation.
  • Multilingual Support: Offers support for over 29 languages, including major global languages.

Use Cases

This model is suitable for applications requiring a highly flexible instruction-following model with reduced content restrictions. Its enhanced capabilities in coding, mathematics, and structured data understanding make it versatile for various tasks, while its decensored nature allows for broader content generation.

Finetunes
1 models