Code Generation on MBPP (test) (Calibration Metrics)
0.4591AccuracyEPICAR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| EPICARModel=Qwen-3-8B2026.01 | 0.4591 | 0.628 | 0.059 | 0.285 | |
| Base ModelModel=Qwen-3-8B2026.01 | 0.4514 | 0.622 | 0.066 | 0.286 | |
| STaRModel=Qwen-3-8B2026.01 | 0.4514 | 0.577 | 0.066 | 0.285 | |
| EPICARModel=Llama-3-8B2026.01 | 0.393 | 0.538 | 0.113 | 0.246 | |
| STaRModel=Llama-3-8B2026.01 | 0.3774 | 0.523 | 0.39 | 0.387 | |
| Base ModelModel=Llama-3-8B2026.01 | 0.3735 | 0.551 | 0.398 | 0.391 |