Multi-turn dialogue on MT-Eval
8.16LLM-EVAL ScoreMDS
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MDSBackbone=LLAMA3-8B-INSTRUCT, Selection Method=MDS2026.04 | 8.16 | 7.52 | 58.4 | 0.857 | 1.25 | |
| MDSBackbone=QWEN3-8B-INSTRUCT, Selection Method=MDS2026.04 | 8.16 | 8.26 | 59.3 | 0.848 | 1.25 | |
| All DataBackbone=LLAMA3-8B-INSTRUCT, Selection Method=All Data2026.04 | 8.09 | 7.41 | 56.7 | 0.845 | 5.67 | |
| SuperFilteringBackbone=LLAMA3-8B-INSTRUCT, Selection Method=SuperFiltering2026.04 | 8.08 | 7.44 | 56.8 | 0.846 | 4.92 | |
| ZIPBackbone=LLAMA3-8B-INSTRUCT, Selection Method=ZIP2026.04 | 8.06 | 7.42 | 57 | 0.845 | 5.58 | |
| DialScoreBackbone=LLAMA3-8B-INSTRUCT, Selection Method=DialScore2026.04 | 8.05 | 7.21 | 56.7 | 0.845 | 6 | |
| CC-ScoreBackbone=LLAMA3-8B-INSTRUCT, Selection Method=CC-Score2026.04 | 8.05 | 7.48 | 57 | 0.845 | 4.58 | |
| CC-ScoreBackbone=QWEN3-8B-INSTRUCT, Selection Method=CC-Score2026.04 | 8.05 | 8.18 | 57.9 | 0.845 | 4.75 | |
| BackboneBackbone=LLAMA3-8B-INSTRUCT, Selection Method=Backbone2026.04 | 8.04 | 7.44 | 56.9 | 0.831 | 5 | |
| RethinkingBackbone=LLAMA3-8B-INSTRUCT, Selection Method=Rethinking2026.04 | 8.01 | 7.43 | 56.8 | 0.845 | 5.5 | |
| SuperFilteringBackbone=QWEN3-8B-INSTRUCT, Selection Method=SuperFiltering2026.04 | 8.01 | 8.2 | 58.1 | 0.847 | 5.33 | |
| Random DataBackbone=LLAMA3-8B-INSTRUCT, Selection Method=Random Data2026.04 | 8 | 7.41 | 55.8 | 0.842 | 6.92 | |
| HeuristicBackbone=QWEN3-8B-INSTRUCT, Selection Method=Heuristic2026.04 | 8 | 8.1 | 55.3 | 0.828 | 5.67 | |
| HeuristicBackbone=LLAMA3-8B-INSTRUCT, Selection Method=Heuristic2026.04 | 7.99 | 7.23 | 56.2 | 0.838 | 7 | |
| ZIPBackbone=QWEN3-8B-INSTRUCT, Selection Method=ZIP2026.04 | 7.98 | 8.12 | 56.4 | 0.841 | 6.17 | |
| Random DataBackbone=QWEN3-8B-INSTRUCT, Selection Method=Random Data2026.04 | 7.96 | 8.05 | 55.8 | 0.844 | 5.58 | |
| DialScoreBackbone=QWEN3-8B-INSTRUCT, Selection Method=DialScore2026.04 | 7.92 | 8.21 | 58.5 | 0.85 | 4.67 | |
| RethinkingBackbone=QWEN3-8B-INSTRUCT, Selection Method=Rethinking2026.04 | 7.91 | 8.08 | 57.5 | 0.846 | 5.75 | |
| All DataBackbone=QWEN3-8B-INSTRUCT, Selection Method=All Data2026.04 | 7.9 | 8.08 | 56.8 | 0.843 | 7.25 | |
| BackboneBackbone=QWEN3-8B-INSTRUCT, Selection Method=Backbone2026.04 | 7.81 | 7.9 | 49.6 | 0.826 | 7.25 |