Classification Probing on Cities (test)
100Probe Accuracy (Best Layer)xRFM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| xRFMBackbone=Qwen-3B-Instruct2026.01 | 100 | 91.4 | |
| RAPTORBackbone=Qwen-3B-Instruct2026.01 | 100 | 87.8 | |
| RAPTORBackbone=Qwen-7B-Instruct2026.01 | 100 | 90.6 | |
| xRFMBackbone=Llama-3.1-8B-Instruct2026.01 | 100 | 96.6 | |
| xRFMBackbone=Llama-3.1-70B-Instruct2026.01 | 100 | 97.1 | |
| xRFMBackbone=Llama-3.3-70B-Instruct2026.01 | 100 | 96.5 | |
| RAPTORBackbone=Llama-3.3-70B-Instruct2026.01 | 100 | 95.3 | |
| GCSBackbone=Qwen-7B-Instruct2026.01 | 99.7 | 87.6 | |
| xRFMBackbone=Qwen-7B-Instruct2026.01 | 99.7 | 93.6 | |
| xRFMBackbone=Qwen-32B-Instruct2026.01 | 99.7 | 95.2 | |
| RAPTORBackbone=Qwen-32B-Instruct2026.01 | 99.7 | 93 | |
| xRFMBackbone=Gemma-7B-it2026.01 | 99.7 | 92.7 | |
| GCSBackbone=Llama-3.1-8B-Instruct2026.01 | 99.7 | 92.5 | |
| RAPTORBackbone=Llama-3.1-8B-Instruct2026.01 | 99.7 | 94.1 | |
| RAPTORBackbone=Llama-3.1-70B-Instruct2026.01 | 99.7 | 95.5 | |
| GCSBackbone=Qwen-32B-Instruct2026.01 | 99.6 | 91.3 | |
| GCSBackbone=Qwen-3B-Instruct2026.01 | 99.5 | 83.7 | |
| GCSBackbone=Llama-3.3-70B-Instruct2026.01 | 99.3 | 94 | |
| GCSBackbone=Llama-3.1-70B-Instruct2026.01 | 99.2 | 93.8 | |
| RAPTORBackbone=Gemma-7B-it2026.01 | 98.7 | 90.2 | |
| GCSBackbone=Gemma-7B-it2026.01 | 98.5 | 87.3 |