Mathematical Reasoning on GSM8K zero-shot (test)
89.46AccuracyEPICAR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| EPICARFamily=Qwen-3-8B, Zero-shot=true2026.01 | 89.46 | 0.722 | 0.364 | 0.216 | |
| EPICARFamily=Qwen-3-4B, Zero-shot=true2026.01 | 88.25 | 0.756 | 0.132 | 0.106 | |
| STaRFamily=Qwen-3-8B, Zero-shot=true2026.01 | 87.95 | 0.678 | 0.215 | 0.147 | |
| STaRFamily=Qwen-3-4B, Zero-shot=true2026.01 | 86.05 | 0.693 | 0.099 | 0.122 | |
| Base ModelFamily=Qwen-3-8B, Zero-shot=true2026.01 | 85.9 | 0.654 | 0.215 | 0.162 | |
| Base ModelFamily=Qwen-3-4B, Zero-shot=true2026.01 | 85.06 | 0.611 | 0.364 | 0.255 | |
| EPICARFamily=Qwen-3-1.7B, Zero-shot=true2026.01 | 79.3 | 0.66 | 0.671 | 0.61 | |
| STaRFamily=Qwen-3-1.7B, Zero-shot=true2026.01 | 78.54 | 0.493 | 0.541 | 0.463 | |
| Base ModelFamily=Qwen-3-1.7B, Zero-shot=true2026.01 | 77.33 | 0.49 | 0.445 | 0.376 | |
| VOYAGER|D_train|=1000, Evaluation Model=Gemma 7B-IT, Inference Mode=Zero-shot2025.12 | 45.7 | — | — | — | |
| VOYAGER|D_train|=500, Evaluation Model=Gemma 7B-IT, Inference Mode=Zero-shot2025.12 | 42.8 | — | — | — | |
| EPICARFamily=Llama-3-8B, Zero-shot=true2026.01 | 37.45 | 0.565 | 0.223 | 0.284 | |
| DEFAULT|D_train|=1000, Evaluation Model=Gemma 7B-IT, Inference Mode=Zero-shot2025.12 | 35.7 | — | — | — | |
| STaRFamily=Llama-3-8B, Zero-shot=true2026.01 | 32.15 | 0.561 | 0.338 | 0.332 | |
| Base ModelFamily=Llama-3-8B, Zero-shot=true2026.01 | 31.39 | 0.565 | 0.329 | 0.324 | |
| EPICARFamily=Llama-3-3B, Zero-shot=true2026.01 | 21.46 | 0.606 | 0.02 | 0.166 | |
| STaRFamily=Llama-3-3B, Zero-shot=true2026.01 | 17.13 | 0.497 | 0.284 | 0.228 | |
| VOYAGER|D_train|=1000, Evaluation Model=Gemma 2B-IT, Inference Mode=Zero-shot2025.12 | 16.4 | — | — | — | |
| Base ModelFamily=Llama-3-3B, Zero-shot=true2026.01 | 14.94 | 0.527 | 0.303 | 0.223 | |
| DEFAULT|D_train|=1000, Evaluation Model=Gemma 2B-IT, Inference Mode=Zero-shot2025.12 | 13.1 | — | — | — | |
| VOYAGER|D_train|=500, Evaluation Model=Gemma 2B-IT, Inference Mode=Zero-shot2025.12 | 12.9 | — | — | — | |
| EPICARFamily=Llama-3-1B, Zero-shot=true2026.01 | 3.64 | 0.645 | 0.922 | 0.887 | |
| Base ModelFamily=Llama-3-1B, Zero-shot=true2026.01 | 3.49 | 0.545 | 0.841 | 0.741 | |
| STaRFamily=Llama-3-1B, Zero-shot=true2026.01 | 3.49 | 0.478 | 0.84 | 0.74 |