Diversity Measurement on AlpacaEval length-matched
0.481D_Can MeanOLMo-2-7B Base
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| OLMo-2-7B BaseStage=Base, n=1502026.06 | 0.481 | 0.038 | — | — | — | |
| OLMo-2-7B SFTStage=SFT, n=1502026.06 | 0.329 | 0.084 | — | — | — | |
| OLMo-2-7B DPOStage=DPO, n=1502026.06 | 0.286 | 0.072 | — | — | — | |
| OLMo-2-7B Instruct (RLVR)Stage=Instruct (RLVR), n=1502026.06 | 0.281 | 0.071 | — | — | — | |
| Base > Instruct (RLVR) ContrastContrast=Base>Instruct (RLVR)2026.06 | — | — | 0.2 | 2.425 | 5.2 | |
| Base > SFT ContrastContrast=Base>SFT2026.06 | — | — | 0.151 | 1.615 | 3.4 | |
| DPO ≠ Instruct (H1') ContrastContrast=DPO ≠ Instruct (H1')2026.06 | — | — | 0.005 | 0.227 | 0.004 | |
| SFT > DPO ContrastContrast=SFT>DPO2026.06 | — | — | 0.044 | 0.675 | 1.7 |