This is a decensored version of google/gemma 4 31B it, made using Heretic v1.2.0 with the Arbitrary Rank Ablation (ARA) method (with row norm preservation) If not sure which GGUF to pick, try gemma 4 31B it heretic ara IQ3 XXS L.gguf (12.6 GB) or gemma 4 31B it heretic ara IQ3 XXS.gguf (12.1 GB) In case of troubles with loading model files, please update to llama.cpp b8751 or newer Abliteration parameters Parameter Value : : : start layer index 30 end layer index 54 preserve good behavior weight 0.1997 steer bad behavior weight 0.0002 overcorrect relative weight 1.0020 neighbor count 11 Performance Metric This model Original model (google/gemma 4 31B it) : : : : : KL divergence 0.0508 0 (by definition) Refusals 11/100 99/100 Hugging Face GitHub Launch Blog Documentation License : Apache 2.0 Authors : Google DeepMind Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on small models) and generating text output. This release includes open weights models in both pre trained and instruction tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support i…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy