Sentence-pair classification on QQP

88.8Accuracy

RLLR

Updated 4d ago

Evaluation Results

Method
RLLR 2024.05	88.8	-
RLLRMIXED 2024.05	88.8	-
RLHF 2024.05	88.7	-
SFT w. rat. 2024.05	88.4	-
RLLR 2024.05	88.2	-
RLLRMIXED 2024.05	88	-
SFT 2024.05	88	-
SFT w. rat. 2024.05	87.9	-
RLHF 2024.05	87.6	-
SFT 2024.05	85.5	-
RLLR 2024.05	84.3	-
RLLRMIXED 2024.05	84.3	-
RLHF 2024.05	84	-
SFT w. rat. 2024.05	83.6	-
SFT 2024.05	83.4	-
RLLR 2024.05	82.6	-
RLLRMIXED 2024.05	82.4	-
RLHF 2024.05	82.2	-
SFT w. rat. 2024.05	81.9	-
SFT 2024.05	81.3	-
BERT (Full Clean Training) 2022.10	0.9141	-
Backdoored Model (No Defense) 2022.10	0.9138	99.98
Backdoored Model (No Defense) 2022.10	0.9097	100
Backdoored Model (No Defense) 2022.10	0.9089	100
Backdoored Model (No Defense) 2022.10	0.8971	100
Fine-pruning 2022.10	0.8564	100
Fine-mixing 2022.10	0.85	56.87
Fine-pruning 2022.10	0.8458	100
Fine-mixing 2022.10	0.8319	69.39
Fine-mixing 2022.10	0.8237	88.71
Fine-mixing 2022.10	0.8232	15.4
Fine-mixing 2022.10	0.8125	38.55
Fine-mixing 2022.10	0.8106	41.14
Fine-mixing 2022.10	0.8093	42.66
Fine-mixing 2022.10	0.8058	46.31
Fine-mixing 2022.10	0.8033	37.75
Initial BERT (Clean fine-tuned) 2022.10	0.772	-
Initial BERT (Clean fine-tuned) 2022.10	0.7337	-
Initial BERT (Clean fine-tuned) 2022.10	0.6978	-
Initial BERT (Clean fine-tuned) 2022.10	0.6495	-