Code Generation on HumanEval (Occ, CVaR, RecD)
48OccFIXED
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FIXEDInference Path=LLaMA-3.1-8B -> LLaMA-3.1-70B, Setting=CMIS2026.04 | 48 | 41 | 243 | |
| RANDOMInference Path=LLaMA-3.1-8B -> LLaMA-3.1-70B, Setting=CMIS2026.04 | 39 | 34 | 187 | |
| PERIODICInference Path=LLaMA-3.1-8B -> LLaMA-3.1-70B, Setting=CMIS2026.04 | 34 | 28 | 166 | |
| OUTPUT-TRIGGERInference Path=LLaMA-3.1-8B -> LLaMA-3.1-70B, Setting=CMIS2026.04 | 26 | 22 | 145 | |
| CONFIDENCE-THRESHOLDInference Path=LLaMA-3.1-8B -> LLaMA-3.1-70B, Setting=CMIS2026.04 | 25 | 21 | 150 | |
| RISK PREDICTORInference Path=LLaMA-3.1-8B -> LLaMA-3.1-70B, Setting=CMIS2026.04 | 23 | 19 | 133 | |
| VEROICInference Path=LLaMA-3.1-8B -> LLaMA-3.1-70B, Setting=CMIS2026.04 | 19 | 17 | 114 |