Classification Probing on Common (test)
76.9Probe Accuracy (Best Layer)xRFM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| xRFMBackbone=Llama-3.3-70B-Instruct2026.01 | 76.9 | 72.5 | |
| xRFMBackbone=Llama-3.1-70B-Instruct2026.01 | 76.8 | 72.5 | |
| RAPTORBackbone=Llama-3.3-70B-Instruct2026.01 | 76.3 | 72 | |
| xRFMBackbone=Qwen-32B-Instruct2026.01 | 75.9 | 71.1 | |
| RAPTORBackbone=Llama-3.1-70B-Instruct2026.01 | 75.8 | 72.3 | |
| RAPTORBackbone=Qwen-32B-Instruct2026.01 | 75.5 | 70.8 | |
| xRFMBackbone=Llama-3.1-8B-Instruct2026.01 | 74.5 | 71.8 | |
| RAPTORBackbone=Llama-3.1-8B-Instruct2026.01 | 74.5 | 71.1 | |
| GCSBackbone=Llama-3.1-70B-Instruct2026.01 | 74.4 | 70.4 | |
| GCSBackbone=Llama-3.3-70B-Instruct2026.01 | 74.3 | 69.6 | |
| xRFMBackbone=Qwen-3B-Instruct2026.01 | 73.7 | 68.8 | |
| RAPTORBackbone=Qwen-7B-Instruct2026.01 | 73.7 | 68.7 | |
| RAPTORBackbone=Gemma-7B-it2026.01 | 73.5 | 68.9 | |
| xRFMBackbone=Qwen-7B-Instruct2026.01 | 73.4 | 69.2 | |
| RAPTORBackbone=Qwen-3B-Instruct2026.01 | 73.3 | 68.3 | |
| GCSBackbone=Qwen-32B-Instruct2026.01 | 73.3 | 68.1 | |
| xRFMBackbone=Gemma-7B-it2026.01 | 72.9 | 68.8 | |
| GCSBackbone=Llama-3.1-8B-Instruct2026.01 | 72 | 67.8 | |
| GCSBackbone=Qwen-7B-Instruct2026.01 | 71.2 | 65.9 | |
| GCSBackbone=Gemma-7B-it2026.01 | 70.2 | 65 | |
| GCSBackbone=Qwen-3B-Instruct2026.01 | 68.7 | 62.5 |