AI & ML interests
Abliteration, interpretability, training-free model modification, democratization of AI
Recent Activity
As of May 21st, 2026, the model heretic-org/Meta-Llama-3.1-8B-Instruct-heretic is no longer available following a legal notice from Meta, Inc.
The official organization of the Heretic project
Heretic is a tool that removes censorship (aka "safety alignment") from transformer-based language models without expensive post-training. It combines an advanced implementation of directional ablation, also known as "abliteration" (Arditi et al. 2024, Lai 2025 (1, 2)), with a TPE-based parameter optimizer powered by Optuna.
The purpose of this organization is to publish and curate high-quality abliterated models made using Heretic.
Membership is open to anyone who has either
- contributed code to the Heretic project, or
- published a well-received model made using Heretic.
To become a member, please join our Discord and request an invitation.
-
heretic-org/Qwen3-4B-Thinking-2507-heretic
Text Generation • 4B • Updated • 440 • 19 -
heretic-org/Qwen3-4B-Instruct-2507-heretic
Text Generation • 4B • Updated • 210 • 14 -
heretic-org/Qwen-3-VL-8B-Instruct-heretic
Image-Text-to-Text • 9B • Updated • 584 • 8 -
heretic-org/Qwen-3-VL-2B-Instruct-heretic
Image-Text-to-Text • 2B • Updated • 1
-
heretic-org/Qwen3-4B-Thinking-2507-heretic
Text Generation • 4B • Updated • 440 • 19 -
heretic-org/Qwen3-4B-Instruct-2507-heretic
Text Generation • 4B • Updated • 210 • 14 -
heretic-org/Qwen-3-VL-8B-Instruct-heretic
Image-Text-to-Text • 9B • Updated • 584 • 8 -
heretic-org/Qwen-3-VL-2B-Instruct-heretic
Image-Text-to-Text • 2B • Updated • 1