This is a decensored version of prism ml/Ternary Bonsai 27B unpacked, made using Heretic v1.4.0 with the Arbitrary Rank Ablation (ARA) method using a LoRA adapter and row norm preservation Abliteration parameters Parameter Value : : : start layer index 10 end layer index 51 preserve good behavior weight 0.6491 steer bad behavior weight 0.0008 overcorrect relative weight 0.9874 neighbor count 14 Performance Metric This model Original model (prism ml/Ternary Bonsai 27B unpacked) : : : : : Keywords 2/100 98/100 KL divergence 0.0243 0 (by definition) Note: Performance testing, including the measurement of refusal rates, was conducted using Japanese datasets (harmless alpaca ja, harmful behaviors ja). Ternary Bonsai 27B — Unpacked FP16 Safetensors FP16 safetensors (HuggingFace format) of the Ternary Bonsai 27B model. This repo exists for users who want to run Ternary Bonsai with stock HuggingFace tooling or frameworks that don't yet support the packed ternary format. The 2 bit hybrid attention kernels are currently in our forks of MLX, mlx swift, and llama.cpp — once they land upstream, this unpacked version will no longer be needed. We strongly recommend using the natively packed model…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy