qwen2.5-3b-instruct-cn-dat-kr0.2-a2.0-creative

qwen2.5-3b-instruct-cn-dat-kr0.2-a2.0-creative

This is a CreativityNeuro (CN) modified version of Qwen/Qwen2.5-3B-Instruct.

Model Details

• Base Model: Qwen/Qwen2.5-3B-Instruct
• Modification: CreativityNeuro weight scaling
• Prompt Set: dat
• Keep Ratio: 0.2 (top 20.0% of task-specific weights)
• Alpha: 2.0 (scaling strength)
• Mode: creative

What is CreativityNeuro?

CreativityNeuro identifies task-specific neurons using Wanda-style importance scoring and selectively upscales weights associated with creative thinking. The modification formula is:

W_new = W × (1 + α × mask)

Where mask identifies weights important for creative tasks but not for routine/associative tasks.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("priorcomputers/qwen2.5-3b-instruct-cn-dat-kr0.2-a2.0-creative")
tokenizer = AutoTokenizer.from_pretrained("priorcomputers/qwen2.5-3b-instruct-cn-dat-kr0.2-a2.0-creative")

# Use like any other model
outputs = model.generate(...)

Citation

If you use this model, please cite:

@misc{creativityneuro2025,
  title={CreativityNeuro: Mechanistic Interpretability for LLM Creativity},
  author={Prior Computers},
  year={2025},
  url={https://huggingface.co/priorcomputers}
}