V3: Refusal direction via projection
A single refusal direction, measured as the difference between harmful and harmless prompts, applied at run time to remove or add refusals. Abliteration shipped as a vector instead of edited weights.
Original source: Arditi et al. 2024, Refusal in Language Models Is Mediated by a Single Direction ↗
Repos
7
Vectors
7
All repos using V3
alesha-pro/Qwen3.8-Flash-Next-abliterated-GSQ-RCO-Strata-GGUF
Source cited: Arditi et al. 2024 'Refusal is Mediated by a Single Direction'; huihui-ai Huihui-Qwen3.8-Flash-Next-abliterated; Strata runtime (Niko1221)
1 vector 1 set · 472 KB
345,445 repo downloads
345,445 repo downloads
alesha-pro/Qwen3.8-Flash-Next-abliterated-GSQ-RCO-Coder-Strata-GGUF
Source cited: Arditi et al. 2024, 'Refusal in Language Models Is Mediated by a Single Direction'; huihui-ai abliteration; Strata runtime
1 vector 1 set · 472 KB
21,840 repo downloads
21,840 repo downloads
alesha-pro/Qwen3.8-27B-S-mirai-GGUF
Source cited: Arditi et al. 2024, 'Refusal is Mediated by a Single Direction'; llama.cpp control vector projection mode (alesha-pro/llama.cpp-mirai-s)
1 vector 1 set · 1.23 MB
16,695 repo downloads
16,695 repo downloads
alesha-pro/Qwen3.8-Flash-Next-S-abliterated-Strata-GGUF
Source cited: Arditi et al. 2024 'Refusal is Mediated by a Single Direction'; direction derived from huihui-ai/Huihui-Qwen3.8-Flash-Next-abliterated; Strata engine
1 vector 1 set · 472 KB
10,868 repo downloads
10,868 repo downloads
Hanselor/Qwen-Flash-abliterated-Strata-GGUF
Source cited: Arditi et al. 2024 (Refusal is Mediated by a Single Direction); huihui-ai abliterated checkpoint; Strata engine (Niko1221)
1 vector 1 set · 472 KB
1,990 repo downloads
1,990 repo downloads
QWER123951/Qwen3.8-Flash-Next-abliterated-GSQ-RCO-Strata-GGUF
Source cited: Arditi et al. 2024 'Refusal is Mediated by a Single Direction'; huihui-ai Huihui-Qwen3.8-Flash-Next-abliterated; Strata runtime (Niko1221)
1 vector 1 set · 472 KB
887 repo downloads
887 repo downloads
tpls/gemma-4-12B-coder-fable5-composer2.5-v1-steer
Source cited: llama.cpp control vector GGUF; no paper cited
1 vector 1 set · 707 KB
81 repo downloads
81 repo downloads