What the fuck is this model about?
This model, blackbook-lm/Qwen2.5-7B-Instruct-heretic, is a decensored version of the original Qwen/Qwen2.5-7B-Instruct model. It was created using the Heretic v1.2.0 tool, specifically to reduce content refusals and provide less restricted output.
What makes THIS different from all the other models?
The primary differentiator is its significantly reduced refusal rate. While the original Qwen2.5-7B-Instruct model had 99 out of 100 refusals in testing, this 'heretic' version shows only 5 out of 100 refusals. This makes it suitable for applications where the base model's content restrictions might be too stringent. It maintains the core capabilities of the Qwen2.5 series, including:
- Enhanced Knowledge & Reasoning: Improved capabilities in coding and mathematics, leveraging specialized expert models.
- Advanced Instruction Following: Better at understanding and executing complex instructions, generating long texts (over 8K tokens), and handling structured data like JSON.
- Robust Long-Context Support: Features a full context length of 131,072 tokens, with generation up to 8,192 tokens, utilizing techniques like YaRN for long text processing.
- Multilingual Support: Capable of processing and generating text in over 29 languages.
Should I use this for my use case?
You should consider using this model if:
- Your application requires a powerful 7.61 billion parameter instruction-tuned model with strong general capabilities in coding, mathematics, and instruction following.
- You need a model that can handle very long contexts (up to 128K tokens).
- Your use case involves multilingual interactions across a broad range of languages.
- Crucially, if the content restrictions or refusal rates of standard instruction-tuned models are hindering your application, this decensored version offers a more permissive alternative while retaining the underlying performance of Qwen2.5.