Classification Probing on STSA (test)
0.955Probe Accuracy (Best Layer)RAPTOR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RAPTORBackbone=Qwen-32B-Instruct2026.01 | 0.955 | 0.906 | |
| xRFMBackbone=Llama-3.1-70B-Instruct2026.01 | 0.955 | 0.923 | |
| xRFMBackbone=Qwen-32B-Instruct2026.01 | 0.953 | 0.903 | |
| RAPTORBackbone=Llama-3.3-70B-Instruct2026.01 | 0.952 | 0.921 | |
| RAPTORBackbone=Qwen-3B-Instruct2026.01 | 0.95 | 0.882 | |
| xRFMBackbone=Llama-3.3-70B-Instruct2026.01 | 0.95 | 0.919 | |
| xRFMBackbone=Qwen-7B-Instruct2026.01 | 0.949 | 0.9 | |
| RAPTORBackbone=Llama-3.1-70B-Instruct2026.01 | 0.949 | 0.924 | |
| RAPTORBackbone=Qwen-7B-Instruct2026.01 | 0.947 | 0.902 | |
| xRFMBackbone=Llama-3.1-8B-Instruct2026.01 | 0.947 | 0.916 | |
| RAPTORBackbone=Llama-3.1-8B-Instruct2026.01 | 0.947 | 0.918 | |
| GCSBackbone=Qwen-32B-Instruct2026.01 | 0.945 | 0.883 | |
| xRFMBackbone=Qwen-3B-Instruct2026.01 | 0.943 | 0.877 | |
| GCSBackbone=Llama-3.1-8B-Instruct2026.01 | 0.939 | 0.899 | |
| GCSBackbone=Llama-3.3-70B-Instruct2026.01 | 0.939 | 0.898 | |
| GCSBackbone=Qwen-7B-Instruct2026.01 | 0.938 | 0.877 | |
| GCSBackbone=Llama-3.1-70B-Instruct2026.01 | 0.938 | 0.902 | |
| GCSBackbone=Qwen-3B-Instruct2026.01 | 0.935 | 0.836 | |
| xRFMBackbone=Gemma-7B-it2026.01 | 0.932 | 0.886 | |
| RAPTORBackbone=Gemma-7B-it2026.01 | 0.93 | 0.889 | |
| GCSBackbone=Gemma-7B-it2026.01 | 0.921 | 0.856 |