List your company

Concept Faithfulness Evaluation on ERASER Movie

90.92Perturbed AccuracySAE
2.540825.485448.4371.3746Jun 24, 2025

Evaluation results

MethodLinks
SAEModel=Llama2025.0690.920.3
SAEModel=Qwen2025.0687.310.7
Single-LayerModel=Llama2025.0680.2811.3
CLVQ-VAEModel=Llama2025.0678.5113.3
ClusteringModel=Llama2025.0677.7914.1
ClusteringModel=BERT2025.0677.576
Single-LayerModel=BERT2025.0676.247.6
ClusteringModel=Qwen2025.0667.422.8
ClusteringModel=RoBERTa2025.0662.7128.6
Single-LayerModel=Qwen2025.0661.4229.6
CLVQ-VAEModel=Qwen2025.0661.1330
CLVQ-VAEModel=BERT2025.0653.1135.6
SAEModel=BERT2025.0646.5743.5
SAEModel=RoBERTa2025.0643.6150.3
Single-LayerModel=RoBERTa2025.066.3392.8
CLVQ-VAEModel=RoBERTa2025.065.9493.2