Fully automatic censorship removal for language models
- Updated
Jul 24, 2026 - Python
Fully automatic censorship removal for language models
Fully uncensored, capability-enhanced abliteration of Qwen3.6-27B. NVFP4 + z-lab DFlash speculative decoding (n=12) on the unified ghcr.io/aeon-7/aeon-vllm-ultimate:latest container, tuned for long-context draft acceptance on DGX Spark. 6 HF variants (BF16/NVFP4/MTP/MTP-XS), docker-compose, and QuickStart.
Jacobian-Brainwash : A manual alignment tool for large language models built on Anthropic's Jacobian Lens. Results are exportable.
Automated alignment adjustment for LLMs — direct steering, LoRA, and MoE expert-granular abliteration, optimized via multi-objective Optuna TPE.
Make abliterated models with transformers, easy and fast
Powerful no-code LLM fine-tuner: upload data → train → deploy in minutes. Unsloth 2-5× acceleration · QLoRA/DPO/RLHF/PPO/ORPO · Reward Model training · GGUF export · vLLM inference · BLEU/ROUGE/BERTScore · full CLI · Heretic Mode to unlock full model potential
Enhanced fork of Heretic (an automated LLM de-censoring tool) optimized for macOS (Apple Silicon) with checkpoint system and LM Studio integration
Gemma 4 31B Abliterated — quality-preserving guardrail removal for Google's most capable open model. Apache 2.0. Runs on Apple Silicon via MLX.
modify a language model's behavior by abliterating its weights.
Advanced abliteration framework: 8-stage pipeline, auto fine-tuning, voice support, real-time collaboration, security scanning | Production-grade LLM liberation
GLM-5.2, completely uncensored and fully local on 4 cards — the think-off recipe plus full serving + reproduction guide.
Layer-by-layer model training and modification for 80B+ MoE models on consumer GPUs. Abliteration, LongRoPE, LoRA merge, weight visualization. Built because nothing else could do it. https://justcalljon.pro
🚀 Train and modify 80B+ parameter Mixture of Experts models layer-by-layer on consumer GPUs using Python with AEGIS AI Trainer.
Local-first AI workstation. Run open-weight models, fine-tune, orchestrate multi-agent teams. No cloud required.
MLX-native toolkit for understanding and reshaping how language models behave on Apple Silicon
Archive for Heretic-generated model reproducibility records. Fully local, single-command setup.
🔓 Ablate — directional ablation (abliteration) toolkit for open-source LLMs. Automatic censorship/refusal removal via residual-stream direction ablation, with KL-guided search, an LLM-judge harness, and one-call push to the Hub. pip install ablate-llm
Generate cybersecurity prompt datasets for Heretic abliteration and AI safety research
Toolkit to abliterate any instruct-tuned LLM using Arditi 2024 + NousResearch method
Uncensoring LLMs via Albiteration and rehabilitating via RLVR/GRPO with small post training corpus