This is a decensored version of google/gemma 4 12B it, made using Heretic v1.3.0 Created and evaluated using custom system prompt. If not sure which GGUF to pick, try gemma 4 12B it heretic custom Q6 K.gguf (9.8 GB) or gemma 4 12B it heretic custom IQ4 NL.gguf (7 GB) [!TIP] This model is reproducible! See the README in the reproduce directory for more information. Abliteration parameters Parameter Value : : : direction index 32.83 attn.o proj.max weight 1.41 attn.o proj.max weight position 28.72 attn.o proj.min weight 0.64 attn.o proj.min weight distance 21.64 mlp.down proj.max weight 1.46 mlp.down proj.max weight position 43.58 mlp.down proj.min weight 1.38 mlp.down proj.min weight distance 15.93 Performance Metric This model Original model (google/gemma 4 12B it) : : : : : KL divergence 0.0162 0 (by definition) Refusals 23/100 98/100 Hugging Face GitHub Launch Blog Documentation License : Apache 2.0 Authors : Google DeepMind [!Note] This model card is for the Gemma 4 12B Unified model, which is part of the Gemma 4 family of open models. Built with the same multimodal functionality as Gemma 4 E2B and E4B (text, audio, image, and video inputs), it brings native audio and vision und…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy