Classification Probing on HateXplain (test)
79.1Probe Accuracy (Best Layer)xRFM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| xRFMBackbone=Llama-3.1-70B-Instruct2026.01 | 79.1 | 76.6 | |
| xRFMBackbone=Llama-3.3-70B-Instruct2026.01 | 79.1 | 76 | |
| RAPTORBackbone=Llama-3.1-70B-Instruct2026.01 | 79 | 77.2 | |
| RAPTORBackbone=Llama-3.3-70B-Instruct2026.01 | 78.5 | 76.6 | |
| RAPTORBackbone=Llama-3.1-8B-Instruct2026.01 | 77.3 | 74.9 | |
| xRFMBackbone=Qwen-32B-Instruct2026.01 | 77.1 | 73.9 | |
| RAPTORBackbone=Qwen-32B-Instruct2026.01 | 77.1 | 74.3 | |
| xRFMBackbone=Llama-3.1-8B-Instruct2026.01 | 76.8 | 74.3 | |
| GCSBackbone=Llama-3.1-70B-Instruct2026.01 | 76.6 | 74.7 | |
| GCSBackbone=Llama-3.3-70B-Instruct2026.01 | 76.2 | 74.3 | |
| xRFMBackbone=Qwen-7B-Instruct2026.01 | 75.8 | 72.5 | |
| RAPTORBackbone=Qwen-7B-Instruct2026.01 | 75.8 | 73.1 | |
| xRFMBackbone=Gemma-7B-it2026.01 | 75.7 | 71.9 | |
| RAPTORBackbone=Gemma-7B-it2026.01 | 75.7 | 72.3 | |
| RAPTORBackbone=Qwen-3B-Instruct2026.01 | 75.2 | 71.1 | |
| xRFMBackbone=Qwen-3B-Instruct2026.01 | 74.3 | 70.1 | |
| GCSBackbone=Qwen-32B-Instruct2026.01 | 74.1 | 71.3 | |
| GCSBackbone=Llama-3.1-8B-Instruct2026.01 | 74 | 71.8 | |
| GCSBackbone=Qwen-7B-Instruct2026.01 | 72.2 | 69.3 | |
| GCSBackbone=Gemma-7B-it2026.01 | 71.4 | 68 | |
| GCSBackbone=Qwen-3B-Instruct2026.01 | 69.7 | 66.1 |