darkc0de/Mistral-Small-3.2-24B-Instruct-2506-heretic
The darkc0de/Mistral-Small-3.2-24B-Instruct-2506-heretic is a 24 billion parameter instruction-tuned language model, a decensored version of Mistral-Small-3.2-24B-Instruct-2506 created using Heretic v1.2.0. It features a 32768 token context length and significantly reduces refusals compared to its original counterpart (4/100 vs 98/100). This model excels in instruction following, reducing repetition errors, and robust function calling, making it suitable for applications requiring less restrictive content generation and improved reliability.
Loading preview...
Model Overview
darkc0de/Mistral-Small-3.2-24B-Instruct-2506-heretic is a 24 billion parameter instruction-tuned language model, derived from Mistral-Small-3.2-24B-Instruct-2506 and decensored using Heretic v1.2.0. It maintains a 32768 token context length and builds upon the Mistral-Small-3.2 improvements.
Key Capabilities
- Decensored Output: Significantly reduces refusals, with only 4 refusals per 100 prompts compared to 98/100 in the original model, offering more unrestricted content generation.
- Enhanced Instruction Following: Demonstrates improved accuracy in following precise instructions, scoring 65.33% on Wildbench v2 and 43.1% on Arena Hard v2.
- Reduced Repetition Errors: Decreases infinite generations by 2x on challenging, long, and repetitive prompts, achieving a 1.29% rate.
- Robust Function Calling: Features a more robust function calling template, supporting complex tool use and vision reasoning.
- Multimodal Capabilities: Supports vision inputs, enabling reasoning based on images, as demonstrated in examples involving image analysis for decision-making.
- Improved STEM Performance: Shows slight improvements in STEM benchmarks, including a 69.06% on MMLU Pro (5-shot CoT) and 92.90% on HumanEval Plus - Pass@5.
Good For
- Applications requiring less restrictive or decensored text generation.
- Tasks demanding high instruction adherence and reduced repetitive outputs.
- Complex function calling and tool use scenarios.
- Multimodal applications involving image understanding and reasoning.
- Developers seeking a powerful 24B parameter model with strong performance in instruction following and code-related tasks.