My mergedheretic model finetuned on my creativeharm dataset. It's better than I expected and surprisingly uncensored. Still needs work though. Uploaded to track my progress. If you want to use this model, may I suggest my HereticFT instead? It's a little better in my opinion but feel free to do whatever you want, really.

This gemma3 model was trained 2x faster with Unsloth and Huggingface's TRL library.