Faithfulness Evaluation on SST2 benchmark leaderboard · SOTA2 Research

Faithfulness Evaluation on SST2

0.563AUC π-Soft (NS)Grad-ELLM

Evaluation Results

Method	Links
Grad-ELLM2026.01		0.563	0.304
Random2026.01		0.549	0.318
Integrated Gradients2026.01		0.526	0.276
Saliency2026.01		0.503	0.253
DeepLIFT2026.01		0.493	0.255
Input×Gradients2026.01		0.493	0.258
Layer Gradient×Activation2026.01		0.493	0.258
Saliency2026.01		0.413	0.347
Saliency2026.01		0.413	0.347
Grad-ELLM2026.01		0.396	0.464
Grad-ELLM2026.01		0.396	0.464
Attention2026.01		0.361	0.165
Integrated Gradients2026.01		0.357	0.366
Integrated Gradients2026.01		0.357	0.366
Random2026.01		0.337	0.56
Random2026.01		0.337	0.56
Value Zeroing2026.01		0.325	0.136
Attention2026.01		0.304	0.2
Attention2026.01		0.304	0.2
DeepLIFT2026.01		0.3	0.278
DeepLIFT2026.01		0.3	0.278
Input×Gradients2026.01		0.297	0.277
Layer Gradient×Activation2026.01		0.297	0.275
Input×Gradients2026.01		0.297	0.277
Layer Gradient×Activation2026.01		0.297	0.275
Value Zeroing2026.01		0.249	0.099
Value Zeroing2026.01		0.249	0.099