HarethahMo/Qwen2.5-1.5B-Instruct-heretic
HarethahMo/Qwen2.5-1.5B-Instruct-heretic is a 1.5 billion parameter instruction-tuned causal language model, based on Qwen2.5, with a 32,768 token context length. This model is a decensored version of the original Qwen/Qwen2.5-1.5B-Instruct, specifically modified to reduce refusals. It retains the Qwen2.5 improvements in coding, mathematics, long text generation, and structured data understanding, making it suitable for applications requiring less restrictive content policies.
Loading preview...
HarethahMo/Qwen2.5-1.5B-Instruct-heretic Overview
This model is a 1.5 billion parameter instruction-tuned causal language model, derived from the Qwen2.5 series. It is a decensored variant of the original Qwen/Qwen2.5-1.5B-Instruct, created using the Heretic v1.1.0 tool. The primary differentiation of this model is its significantly reduced refusal rate, with only 3 refusals out of 100 compared to 99/100 for the original model, as indicated by KL divergence metrics.
Key Capabilities Inherited from Qwen2.5:
- Enhanced Knowledge & Specialized Skills: Improved performance in coding and mathematics, leveraging specialized expert models.
- Instruction Following & Text Generation: Significant advancements in adhering to instructions, generating long texts (up to 8K tokens), and understanding/producing structured data like JSON.
- Robustness: More resilient to diverse system prompts, improving role-play and chatbot condition-setting.
- Long-Context Support: Supports a context length of up to 32,768 tokens, with generation capabilities up to 8,192 tokens.
- Multilingual Support: Comprehensive support for over 29 languages, including major global languages.
Ideal Use Cases:
- Applications requiring a language model with fewer content restrictions or censorship.
- Tasks benefiting from strong coding, mathematical, and structured data processing abilities.
- Scenarios demanding long-context understanding and generation.
- Multilingual applications where broad language support is crucial.