List your company
ResearchFormerly WizWand

Language Model Unlearning on TOFU (Forget10)

100Forget Quality (FQ)Retain90
-4235077Sep 29, 2025Nov 3, 2025Dec 9, 2025Jan 14, 2026Feb 18, 2026Mar 26, 2026May 1, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Retain90Backbone=Phi-1.5B2025.09100524391—
GD+SineBackbone=Phi-1.5B, Method Category=Parameter-Efficient Methods, Params (%)=1.62025.0994.3522290—
ME+GD (LoRA)Backbone=Phi-1.5B, Method Category=Parameter-Efficient Methods, Params (%)=1.62025.0978.6521493—
GD+TanhBackbone=Phi-1.5B, Method Category=Parameter-Efficient Methods, Params (%)=1.62025.0934.2492885—
before unlearningBackbone Model=Qwen-3-8B2026.050.902576.03——88.47
before unlearningBackbone Model=Llama-2-7B2026.050.812262.19——83.64
before unlearningBackbone Model=Phi-1.52026.050.579652.88——43.17
NPOBackbone Model=Llama-2-7B, Unlearning Granularity=Sequence-Level2026.050.213960.05——47.12
GABackbone Model=Qwen-3-8B, Unlearning Granularity=Sequence-Level2026.050.173957.3——45.42
RMUBackbone Model=Llama-2-7B, Unlearning Granularity=Sequence-Level2026.050.171551.74——29.65
NPOBackbone Model=Qwen-3-8B, Unlearning Granularity=Sequence-Level2026.050.168455.97——49.3
NPOBackbone Model=Phi-1.5, Unlearning Granularity=Sequence-Level2026.050.163647.92——24.2
GABackbone Model=Llama-2-7B, Unlearning Granularity=Sequence-Level2026.050.163250.74——41.79
GABackbone Model=Phi-1.5, Unlearning Granularity=Sequence-Level2026.050.142532.64——19.63
RMUBackbone Model=Phi-1.5, Unlearning Granularity=Sequence-Level2026.050.140449.27——23.75
S-GABackbone Model=Phi-1.5, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.136834.58——28.05
RMUBackbone Model=Qwen-3-8B, Unlearning Granularity=Sequence-Level2026.050.136558.82——40.35
WGABackbone Model=Phi-1.5, Unlearning Granularity=Sequence-Level2026.050.132850.86——34.18
S-NPOBackbone Model=Qwen-3-8B, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.132356.37——61.39
S-RMUBackbone Model=Qwen-3-8B, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.129660.42——54.54
T-RMUBackbone Model=Phi-1.5, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.127351.06——27.19
S-RMUBackbone Model=Phi-1.5, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.125850.97——24.39
T-GABackbone Model=Phi-1.5, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.125535.29——28.47
WGABackbone Model=Llama-2-7B, Unlearning Granularity=Sequence-Level2026.050.125564.39——64.18
S-WGABackbone Model=Phi-1.5, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.124352.37——35.12
S-GABackbone Model=Qwen-3-8B, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.121660.65——65.03
T-GABackbone Model=Qwen-3-8B, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.120963.42——66.23
S-WGABackbone Model=Qwen-3-8B, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.11560.12——65.74
T-NPOBackbone Model=Qwen-3-8B, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.113858.12——63.89
WGABackbone Model=Qwen-3-8B, Unlearning Granularity=Sequence-Level2026.050.107258.24——57.61
S-NPOBackbone Model=Phi-1.5, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.102348.21——27.65
T-NPOBackbone Model=Phi-1.5, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.100449.36——29.85
S-WGABackbone Model=Llama-2-7B, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.095565.12——65.24
T-WGABackbone Model=Llama-2-7B, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.094966.5——67.91
T-WGABackbone Model=Phi-1.5, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.092352.42——36.14
S-GABackbone Model=Llama-2-7B, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.091753.21——57.22
S-RMUBackbone Model=Llama-2-7B, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.083559.14——41.96
T-NPOBackbone Model=Llama-2-7B, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.081760.89——65.83
T-RMUBackbone Model=Qwen-3-8B, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.081459.36——58.39
T-WGABackbone Model=Qwen-3-8B, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.072360.53——68.57
T-GABackbone Model=Llama-2-7B, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.071557.13——63
T-RMUBackbone Model=Llama-2-7B, Unlearning Granularity=Token-Level, Selection Strategy=Hard Selection2026.050.070958.62——43.38
S-NPOBackbone Model=Llama-2-7B, Unlearning Granularity=Token-Level, Selection Strategy=Soft Weighting2026.050.066455.39——59.75
NPOBackbone=Phi-1.5B, Method Category=Full Fine-tuning Methods, Params (%)=100.02025.090.0026374545—
GD+FILABackbone=Phi-1.5B, Method Category=Parameter-Efficient Methods, Params (%)=1.62025.090.000201211—
GDBackbone=Phi-1.5B, Method Category=Full Fine-tuning Methods, Params (%)=100.02025.090363741—
LoKUBackbone=Phi-1.5B, Method Category=Parameter-Efficient Methods, Params (%)=1.62025.090512675—
GABackbone=Phi-1.5B, Method Category=Full Fine-tuning Methods, Params (%)=100.02025.090011—
KLBackbone=Phi-1.5B, Method Category=Full Fine-tuning Methods, Params (%)=100.02025.090011—
GD+LoRABackbone=Phi-1.5B, Method Category=Parameter-Efficient Methods, Params (%)=1.62025.090288545—
DPOBackbone=Phi-1.5B, Method Category=Full Fine-tuning Methods, Params (%)=100.02025.090484167—
GA+FILABackbone=Phi-1.5B, Method Category=Parameter-Efficient Methods, Params (%)=1.62025.090000—
IHLBackbone=Phi-1.5B, Method Category=Full Fine-tuning Methods, Params (%)=100.02025.090515376—
OriginalBackbone=Phi-1.5B2025.090529392—