Concept Monitoring on 11 concept monitoring tasks (test)
99.39AccuracyRFM
Evaluation Results
| Method | Links | |
|---|---|---|
| RFMAggregation Method=Highest accuracy block, Base Model=Llama-3.1-70B2026.04 | 99.39 | |
| RFMAggregation Method=Highest accuracy block, Base Model=Llama-3.1-8B2026.04 | 99.09 | |
| RFMAggregation Method=Highest accuracy block, Base Model=Qwen2.5-7B2026.04 | 98.93 | |
| RFMAggregation Method=Highest accuracy block, Base Model=Llama-3.2-1B2026.04 | 97.9 | |
| LoRAAggregation Method=Highest accuracy block, Base Model=Llama-3.1-70B2026.04 | 97.37 | |
| RFMAggregation Method=Average over all blocks, Base Model=Llama-3.1-70B2026.04 | 97.19 | |
| ReFTAggregation Method=Highest accuracy block, Base Model=Llama-3.1-70B2026.04 | 96.19 | |
| RFMAggregation Method=Average over all blocks, Base Model=Qwen2.5-7B2026.04 | 96.14 | |
| RFMAggregation Method=Average over all blocks, Base Model=Llama-3.1-8B2026.04 | 95.13 | |
| LoRAAggregation Method=Highest accuracy block, Base Model=Qwen2.5-7B2026.04 | 94.86 | |
| ReFTAggregation Method=Highest accuracy block, Base Model=Qwen2.5-7B2026.04 | 94.6 | |
| LoRAAggregation Method=Highest accuracy block, Base Model=Llama-3.1-8B2026.04 | 93.26 | |
| ReFTAggregation Method=Highest accuracy block, Base Model=Llama-3.1-8B2026.04 | 93.16 | |
| RFMAggregation Method=Average over all blocks, Base Model=Llama-3.2-1B2026.04 | 93.02 | |
| LoRAAggregation Method=Highest accuracy block, Base Model=Llama-3.2-1B2026.04 | 89.57 | |
| ReFTAggregation Method=Highest accuracy block, Base Model=Llama-3.2-1B2026.04 | 88.79 | |
| LoRAAggregation Method=Average over all blocks, Base Model=Llama-3.1-70B2026.04 | 74.41 | |
| ReFTAggregation Method=Average over all blocks, Base Model=Llama-3.1-70B2026.04 | 73.77 | |
| LoRAAggregation Method=Average over all blocks, Base Model=Llama-3.1-8B2026.04 | 71.89 | |
| LoRAAggregation Method=Average over all blocks, Base Model=Qwen2.5-7B2026.04 | 70.63 | |
| LoRAAggregation Method=Average over all blocks, Base Model=Llama-3.2-1B2026.04 | 69.81 | |
| ReFTAggregation Method=Average over all blocks, Base Model=Qwen2.5-7B2026.04 | 68.85 | |
| ReFTAggregation Method=Average over all blocks, Base Model=Llama-3.1-8B2026.04 | 68.75 | |
| ReFTAggregation Method=Average over all blocks, Base Model=Llama-3.2-1B2026.04 | 68.23 |