List your company

Toxicity Mitigation on REALTOXICITYPROMPTS

21.24ToxicityEIGENSHIFT + Targeted Subspace Intervention
19.893628.981838.0747.1582Feb 6, 2026

Evaluation results

MethodLinks
EIGENSHIFT + Targeted Subspace InterventionBackbone=Mistral-7B-SFT2026.0221.24
DETOXBackbone=GPT-2 Medium2026.0224.49
DETOX + Targeted Subspace InterventionBackbone=GPT-2 Medium2026.0224.61
EIGENSHIFTBackbone=Mistral-7B-SFT2026.0224.64
Vanilla + Targeted Subspace InterventionBackbone=Mistral-7B-SFT2026.0225.94
DETOX + Targeted Subspace InterventionBackbone=Mistral-7B-SFT2026.0226.24
EIGENSHIFT + Targeted Subspace InterventionBackbone=Mistral-7B2026.0228.3
DETOXBackbone=Mistral-7B-SFT2026.0228.73
DETOX + Targeted Subspace InterventionBackbone=Mistral-7B2026.0229.68
Vanilla + Targeted Subspace InterventionBackbone=Mistral-7B2026.0231.39
DETOX + Targeted Subspace InterventionBackbone=GPT-J-6B2026.0233.99
EIGENSHIFTBackbone=Mistral-7B2026.0234.6
DETOXBackbone=Mistral-7B2026.0234.78
Vanilla + Targeted Subspace InterventionBackbone=GPT-J-6B2026.0236.21
EIGENSHIFT + Targeted Subspace InterventionBackbone=GPT-J-6B2026.0238.41
VanillaBackbone=Mistral-7B-SFT2026.0239.29
DETOXBackbone=GPT-J-6B2026.0242.33
EIGENSHIFTBackbone=GPT-J-6B2026.0244.82
VanillaBackbone=Mistral-7B2026.0250.39
VanillaBackbone=GPT-J-6B2026.0250.89
Vanilla + Targeted Subspace InterventionBackbone=GPT-2 Medium2026.0253.93
EIGENSHIFT + Targeted Subspace InterventionBackbone=GPT-2 Medium2026.0254.06
VanillaBackbone=GPT-2 Medium2026.0254.25
EIGENSHIFTBackbone=GPT-2 Medium2026.0254.9