Gemma 3 12B IT
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 1
- Heretic automates refusal ablation in language models
Heretic is an open-source command-line tool that applies directional ablation to transformer-based language models, combining it with an Optuna-powered TPE parameter optimizer. The project says it jointly reduces refusals and KL divergence from the original model, and supports most dense models, many multimodal and MoE architectures, and some hybrid models such as Qwen3.5. Its author reports 3 refusals across 100 harmful prompts and 0.16 KL divergence for an automated Gemma 3 12B variant, while noting that automated metrics do not replace human evaluation and may vary by platform and hardware.
GitHub Trending · Python · 🔥 0
Experience and discussion from the community
Share my Gemma 3 12B IT experienceAsk about Gemma 3 12B IT
Nobody has shared their experience with Gemma 3 12B IT yet.