Advanced Reasoning Suite on Open LLM Leaderboard (MMLU-PRO, BBH, GPQA, MATH, IFEval) (test)
28.38MMLU-PROCal-DPO
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Cal-DPOBackbone=zephyr-7b-sft-full, Alignment Training Dataset=UltraFeedback Binarized, Evaluation Harness Version=v0.4.02024.12 | 28.38 | 21.72 | 43.55 | 29.78 | 2.42 | 34.87 | 63.23 | |
| zephyr-7b-sft-fullBackbone=zephyr-7b-sft-full, Alignment Training Dataset=None (SFT Baseline), Evaluation Harness Version=v0.4.02024.12 | 27.64 | 3.21 | 41.09 | 29.36 | 2.04 | 28.13 | 58.28 | |
| CPOBackbone=zephyr-7b-sft-full, Alignment Training Dataset=UltraFeedback Binarized, Evaluation Harness Version=v0.4.02024.12 | 27.04 | 13.32 | 42.05 | 28.45 | 2.15 | 33.06 | 57 | |
| DPOBackbone=zephyr-7b-sft-full, Alignment Training Dataset=UltraFeedback Binarized, Evaluation Harness Version=v0.4.02024.12 | 26.73 | 10.49 | 43.27 | 28.44 | 1.36 | 21.76 | 61.26 | |
| SLiCBackbone=zephyr-7b-sft-full, Alignment Training Dataset=UltraFeedback Binarized, Evaluation Harness Version=v0.4.02024.12 | 26.52 | 12.45 | 42.33 | 27.93 | 1.38 | 33.74 | 55.38 | |
| f-DPOBackbone=zephyr-7b-sft-full, Alignment Training Dataset=UltraFeedback Binarized, Evaluation Harness Version=v0.4.02024.12 | 25.96 | 11.05 | 42.39 | 28.05 | 1.27 | 23.18 | 62.01 | |
| IPOBackbone=zephyr-7b-sft-full, Alignment Training Dataset=UltraFeedback Binarized, Evaluation Harness Version=v0.4.02024.12 | 25.87 | 11.52 | 40.59 | 28.15 | 1.25 | 27.14 | 60.84 |