Multi-Category Quality Scoring on MT-Bench
8.03Overall ScoreAdaDPO
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| AdaDPOConfiguration Selection=Best across 16 (lr, beta) combinations2026.05 | 8.03 | — | — | — | — | — | — | — | — | |
| DPOConfiguration Selection=Best across 16 (lr, beta) combinations2026.05 | 8.01 | — | — | — | — | — | — | — | — | |
| SFTConfiguration Selection=Best across 16 (lr, beta) combinations2026.05 | 7.58 | — | — | — | — | — | — | — | — | |
| 16-bit BaselineData Format=16-bit Baseline, Model=LLaMA2-Chat-7B2024.11 | 6.38 | 9.25 | 7.2 | 4.65 | 2.55 | 3.3 | 5.55 | 8.93 | 9.58 | |
| AMXFP4Data Format=AMXFP4, Model=LLaMA2-Chat-7B2024.11 | 5.73 | 8.2 | 5.98 | 4.5 | 2.5 | 3.05 | 5.16 | 7.7 | 8.7 | |
| MXFP4Data Format=MXFP4, Model=LLaMA2-Chat-7B2024.11 | 5.25 | 7.2 | 7.03 | 3.95 | 1.7 | 1.7 | 4.35 | 7.53 | 8.53 | |
| MXFP4-PoTData Format=MXFP4-PoT, Model=LLaMA2-Chat-7B2024.11 | 3.22 | 4.3 | 4.05 | 2.35 | 1.9 | 1.25 | 1.55 | 5.23 | 5.15 |