base_model: meta-llama/Llama-3.1-8B-Instruct
library_name: llama.cpp
tags:
- abliterated
- uncensored
- gguf
- llama-3.1
- text-generation
quantized_by: lecheyne24
llama-3.1-8b-abliterated (Q4_K_M, GGUF)
Uncensored build of meta-llama/Llama-3.1-8B-Instruct.
- Method: refusal-direction rank-1 norm-preserving biprojection (o_proj + down_proj), alpha 1.25, all layers.
- Quant: Q4_K_M
- Files:
llama-3.1-8b-unbound.Q4_K_M.gguf,Meta-Llama-3.1-8B-Instruct-abliterated-Q4_K_M.gguf - Plain-answer gauntlet: 7/8. Only miss: rule-vs-reality answered the bare verdict "Jogging is running." without the marker phrase.
Run
llama-server -m llama-3.1-8b-unbound.Q4_K_M.gguf --port 8083 -c 4096 --alias llama-3.1-8b