Code Generation on HumanEval (Pass@1 and Pass@10)
62Pass@1EXPERT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| EXPERTBase Model=OLMo-3-7B, Setting=RL Zero, Fine-tuning=RLVR2026.04 | 62 | 85.3 | |
| TSVBase Model=OLMo-3-7B, Setting=RL Zero, Fine-tuning=RLVR, Scaling Factor (alpha)=12026.04 | 59 | 82.4 | |
| ACTMatBase Model=OLMo-3-7B, Setting=RL Zero, Fine-tuning=RLVR2026.04 | 58.2 | 83.7 | |
| TABase Model=OLMo-3-7B, Setting=RL Zero, Fine-tuning=RLVR, Scaling Factor (alpha)=0.42026.04 | 57.2 | 85.1 | |
| AVERAGEBase Model=OLMo-3-7B, Setting=RL Zero, Fine-tuning=RLVR2026.04 | 56.8 | 85.1 | |
| ISO-CBase Model=OLMo-3-7B, Setting=RL Zero, Fine-tuning=RLVR, Scaling Factor (alpha)=12026.04 | 55.9 | 84.8 | |
| REGMEANBase Model=OLMo-3-7B, Setting=RL Zero, Fine-tuning=RLVR2026.04 | 55.3 | 81.1 | |
| ZERO-SHOTBase Model=OLMo-3-7B, Setting=RL Zero, Fine-tuning=RLVR2026.04 | 50.8 | 82.2 |