Sentiment Prediction on IMDB (test)
0.014HS-MSERollout over layers
Evaluation results
| Method | Links | |
|---|---|---|
| Rollout over layersLLM=Gemma3-270M, component=Attn, approach=Rollout2026.01 | 0.014 | |
| Rollout over layersLLM=Gemma3-270M, component=W-Attn, approach=Rollout2026.01 | 0.014 | |
| Rollout over layersLLM=Gemma3-270M, component=GlbEnc, approach=Rollout2026.01 | 0.017 | |
| Rollout over layersLLM=Gemma3-270M, component=W-AttnResLN, approach=Rollout2026.01 | 0.019 | |
| Mean over heads & layersLLM=Gemma3-270M, component=W-AttnResLN, approach=Mean2026.01 | 0.02 | |
| Mean over heads & layersLLM=Gemma3-270M, component=GlbEnc, approach=Mean2026.01 | 0.021 | |
| Mean over heads & layersLLM=Gemma3-270M, component=Attn, approach=Mean2026.01 | 0.023 | |
| Mean over heads & layersLLM=Gemma3-270M, component=W-Attn, approach=Mean2026.01 | 0.024 | |
| TensorLensLLM=Gemma3-270M, component=Norm, approach=Tensor2026.01 | 0.029 | |
| TensorLensLLM=Gemma3-270M, component=In+Out, approach=Tensor2026.01 | 0.049 | |
| Rollout over layersLLM=ModernBert-Base, component=Attn, approach=Rollout2026.01 | 0.05 | |
| Rollout over layersLLM=ModernBert-Base, component=GlbEnc, approach=Rollout2026.01 | 0.063 | |
| Rollout over layersLLM=ModernBert-Base, component=W-AttnResLN, approach=Rollout2026.01 | 0.064 | |
| Rollout over layersLLM=ModernBert-Base, component=W-Attn, approach=Rollout2026.01 | 0.066 | |
| Mean over heads & layersLLM=ModernBert-Base, component=Attn, approach=Mean2026.01 | 0.069 | |
| Mean over heads & layersLLM=ModernBert-Base, component=W-Attn, approach=Mean2026.01 | 0.072 | |
| Mean over heads & layersLLM=ModernBert-Base, component=GlbEnc, approach=Mean2026.01 | 0.075 | |
| Mean over heads & layersLLM=ModernBert-Base, component=W-AttnResLN, approach=Mean2026.01 | 0.077 | |
| TensorLensLLM=ModernBert-Base, component=Norm, approach=Tensor2026.01 | 0.081 | |
| TensorLensLLM=ModernBert-Base, component=In+Out, approach=Tensor2026.01 | 0.112 |