Model Card for Model ID

Abliterated model based on qwen/Qwen-2.5-Coder-3B

The model isn't uncensored in the traditional sense but its ability to refuse has been abliterated (ie ablated/obliterated) This is derived from FailSpy and Arditi's Cookbooks with some minor modifications. You can read more about abliterated models on Maxime Labonne's post about them here: https://huggingface.co/blog/mlabonne/abliteration

Model Details

Base model is Qwen2.5-Coder-3B

Model Description

Developed by: C-Nocturnum
Model type: Qwen2.5-3B

Bias, Risks, and Limitations

Ablated model, so no guarantees.

Recommendations

This is a relatively small, abliterated coder model. Fairly stable and sans major hallucination.

Training Data

Refusal direction identified using standard datasets: mlabonne/harmful_behaviors & mlabonne/harmless_alpaca

Citation

Publications pending....