Algebraic Reasoning on AQuA-RAT
66.8AccuracyReFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ReFTBase Model=Qwen 3 4B, Params (%)=0.04%, Rank=82026.02 | 66.8 | 2 | |
| LoRABase Model=Qwen 3 4B, Params (%)=0.41%, Rank=82026.02 | 66.7 | 1.9 | |
| Post-block steeringBase Model=Qwen 3 4B, Params (%)=0.04%, Rank=82026.02 | 65 | 0.2 | |
| SFTBase Model=Qwen 3 4B, Params (%)=100%2026.02 | 64.8 | 0 | |
| SFTBase Model=Llama-3.1-8B, Params (%)=100%2026.02 | 47.7 | 0 | |
| Post-block steeringBase Model=Llama-3.1-8B, Params (%)=0.02%, Rank=82026.02 | 47.6 | 0.1 | |
| LoRABase Model=Llama-3.1-8B, Params (%)=0.26%, Rank=82026.02 | 45.4 | -2.3 | |
| ReFTBase Model=Llama-3.1-8B, Params (%)=0.02%, Rank=82026.02 | 45.4 | -2.3 | |
| Post-block steeringBase Model=Llama-3.2-1B, Params (%)=0.04%, Rank=82026.02 | 36.5 | 0.3 | |
| SFTBase Model=Llama-3.2-1B, Params (%)=100%2026.02 | 36.2 | 0 | |
| LoRABase Model=Llama-3.2-1B, Params (%)=0.45%, Rank=82026.02 | 33.6 | -2.9 | |
| SFTBase Model=gemma-3-1b, Params (%)=100%2026.02 | 32.7 | 0 | |
| Post-block steeringBase Model=gemma-3-1b, Params (%)=0.04%, Rank=82026.02 | 32.1 | -0.6 | |
| LoRABase Model=gemma-3-1b, Params (%)=0.45%, Rank=82026.02 | 31.6 | -1.1 | |
| ReFTBase Model=Llama-3.2-1B, Params (%)=0.04%, Rank=82026.02 | 30.2 | -6 | |
| ReFTBase Model=gemma-3-1b, Params (%)=0.04%, Rank=82026.02 | 24.6 | -8.1 |