LLM Steering on AxBench
0.74Steering ScoreUNISTEER
Evaluation Results
| Method | Links | |
|---|---|---|
| UNISTEERTarget LLM=Qwen2.5-1.5B2026.05 | 0.74 | |
| UNISTEERTarget LLM=Qwen2.5-7B2026.05 | 0.68 | |
| LoReFTTarget LLM=Qwen2.5-7B2026.05 | 0.6 | |
| LoReFTTarget LLM=Qwen2.5-1.5B2026.05 | 0.56 | |
| LoReFTTarget LLM=Llama-3.2-1B2026.05 | 0.42 | |
| UNISTEERTarget LLM=Llama-3.2-1B2026.05 | 0.36 | |
| CAATarget LLM=Qwen2.5-1.5B2026.05 | 0.14 | |
| CAATarget LLM=Qwen2.5-7B2026.05 | 0.11 | |
| RepETarget LLM=Qwen2.5-7B2026.05 | 0.11 | |
| ODESteerTarget LLM=Qwen2.5-1.5B2026.05 | 0.1 | |
| RepETarget LLM=Llama-3.2-1B2026.05 | 0.08 | |
| ODESteerTarget LLM=Llama-3.2-1B2026.05 | 0.08 | |
| ODESteerTarget LLM=Qwen2.5-7B2026.05 | 0.08 | |
| RepETarget LLM=Qwen2.5-1.5B2026.05 | 0.05 | |
| CAATarget LLM=Llama-3.2-1B2026.05 | 0.04 | |
| OriginalTarget LLM=Llama-3.2-1B2026.05 | 0 | |
| OriginalTarget LLM=Qwen2.5-1.5B2026.05 | 0 | |
| OriginalTarget LLM=Qwen2.5-7B2026.05 | 0 |