Downstream Task Evaluation on OpenLLM Leaderboard v1 (test)
63.95MMLU (5-shot)SelectiveDPO
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SelectiveDPOBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 63.95 | 53.94 | 64.76 | 61.5 | 52.5 | 76.1 | 62.1 | |
| SFTBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 63.78 | 45.24 | 61.3 | 56.16 | 47.5 | 76.18 | 58.4 | |
| CPOBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 63.77 | 54.32 | 61.67 | 57.54 | 54.5 | 76.97 | 61.5 | |
| DPOBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 63.37 | 53.46 | 64.78 | 61.67 | 52.5 | 77.05 | 62.1 | |
| KTOBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 63.36 | 55.66 | 64.14 | 60.72 | 55.5 | 76.25 | 62.6 | |
| MixDPOBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 63.2 | 55.49 | 64.78 | 61.58 | 54 | 77.45 | 62.8 | |
| SimPOBase model=LLaMA-3-8B, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 63.11 | 59.39 | 62.3 | 62.27 | 51.5 | 77.21 | 62.6 | |
| SFTBase model=Mistral-7B-v0.1, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 59.77 | 42.86 | 61.91 | 54.95 | 38.5 | 76.89 | 55.8 | |
| KTOBase model=Mistral-7B-v0.1, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 59.73 | 56.51 | 65.18 | 59.43 | 39 | 78.09 | 59.7 | |
| MixDPOBase model=Mistral-7B-v0.1, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 59.73 | 52.07 | 65.78 | 60.21 | 38.5 | 77.77 | 59 | |
| SelectiveDPOBase model=Mistral-7B-v0.1, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 59.08 | 45.97 | 65.12 | 60.38 | 28.5 | 77.37 | 56.1 | |
| SimPOBase model=Mistral-7B-v0.1, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 58.49 | 50.68 | 63.89 | 59.26 | 35.5 | 78.41 | 57.7 | |
| CPOBase model=Mistral-7B-v0.1, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 58.12 | 46.93 | 60.33 | 52.28 | 35.5 | 77.29 | 55.1 | |
| DPOBase model=Mistral-7B-v0.1, Preference dataset=Ultrafeedback, Evaluation repository=lm-evaluation-harness2026.02 | 57.57 | 53.14 | 64.34 | 57.19 | 30.5 | 78.33 | 56.8 |