List your company
ResearchFormerly WizWand

Machine Unlearning on TOFU World Fact 1.0

0.884ROUGE-LBase
0.321360.467430.61350.75957Oct 23, 2024Jan 26, 2025May 1, 2025Aug 5, 2025Nov 8, 2025Feb 11, 2026May 18, 2026
Updated 5mo ago

Evaluation results

MethodLinks
BaseBackbone=Dream-7B-Instruct2026.050.8840.089—
LoRABase Model=LLAMA2-7B-TOFU2024.100.880.460.6
MISTRAL-7B-TOFUModel Type=Base (No Unlearning)2024.100.880.620.78
LLAMA2-7B-TOFUModel Type=Base (No Unlearning)2024.100.870.430.56
LoRABase Model=MISTRAL-7B-TOFU2024.100.870.640.78
GABackbone=LLaDA-8B-Instruct2026.050.8650.156—
LLAMA2-13B-TOFUModel Type=Base (No Unlearning)2024.100.860.460.62
LoRABase Model=LLAMA2-13B-TOFU2024.100.860.480.63
CMCBase Model=MISTRAL-7B-TOFU2024.100.860.630.77
CMCBase Model=LLAMA2-7B-TOFU2024.100.850.450.57
MDUBackbone=LLaDA-8B-Instruct, τ=0.002026.050.8480.193—
GDBackbone=LLaDA-8B-Instruct2026.050.8450.164—
NPOBackbone=LLaDA-8B-Instruct2026.050.8440.145—
MDUBackbone=LLaDA-8B-Instruct, τ=0.252026.050.8420.204—
MDUBackbone=LLaDA-8B-Instruct, τ=0.502026.050.8420.205—
SimNPOBackbone=LLaDA-8B-Instruct2026.050.8360.164—
MDUBackbone=Dream-7B-Instruct, τ=0.002026.050.8360.07—
BaseBackbone=LLaDA-8B-Instruct2026.050.8350.143—
DPOBackbone=LLaDA-8B-Instruct2026.050.8340.125—
δ-UNLEARNINGBase Model=LLAMA2-13B-TOFU2024.100.830.410.59
CMCBase Model=LLAMA2-13B-TOFU2024.100.830.470.65
DPOBackbone=Dream-7B-Instruct2026.050.830.071—
SimNPOBackbone=Dream-7B-Instruct2026.050.8280.075—
MDUBackbone=LLaDA-8B-Instruct, τ=0.752026.050.8270.233—
NPOBackbone=Dream-7B-Instruct2026.050.8260.062—
WGABackbone=Dream-7B-Instruct2026.050.8250.062—
WGABackbone=LLaDA-8B-Instruct2026.050.8150.157—
MDUBackbone=Dream-7B-Instruct, τ=0.252026.050.8110.092—
MDUBackbone=Dream-7B-Instruct, τ=0.502026.050.7960.086—
MDUBackbone=Dream-7B-Instruct, τ=1.002026.050.7910.066—
MDUBackbone=LLaDA-8B-Instruct, τ=1.002026.050.7770.187—
MDUBackbone=Dream-7B-Instruct, τ=0.752026.050.7470.045—
GDBackbone=Dream-7B-Instruct2026.050.7190.077—
GABackbone=Dream-7B-Instruct2026.050.3430.019—