Long-context understanding on LongBench v2 (test)
35.2Overall ScoreSoLo-DPO
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SoLo-DPOBackbone=Qwen2.5-7B-Instruct, Mode=SoLo2025.05 | 35.2 | 37.5 | 33.8 | 39.3 | 31.8 | 35 | |
| SoLo-ORPOBackbone=Llama3.1-8B-Instruct, Mode=SoLo2025.05 | 34.7 | 37.5 | 33 | 40 | 32.1 | 31.2 | |
| ORPOBackbone=Llama3.1-8B-Instruct, Mode=short2025.05 | 33.7 | 36.8 | 31.8 | 41.2 | 29.8 | 28.7 | |
| LongPO(pub)Backbone=Qwen2.5-7B-Instruct2025.05 | 33.3 | 35 | 32 | 40.5 | 30 | 27.8 | |
| SoLo-ORPOBackbone=Qwen2.5-7B-Instruct, Mode=SoLo2025.05 | 33.2 | 36.3 | 31.2 | 39.7 | 28.8 | 30.9 | |
| InstructBackbone=Llama3.1-8B-Instruct2025.05 | 32.5 | 35.5 | 30.7 | 40.7 | 27.7 | 28.5 | |
| ORPOBackbone=Llama3.1-8B-Instruct, Mode=long2025.05 | 32.1 | 34.5 | 30.6 | 39.7 | 27.8 | 28 | |
| SFTBackbone=Llama3.1-8B-Instruct, Mode=short2025.05 | 31.8 | 34.5 | 30 | 38.3 | 28.6 | 27 | |
| SoLo-SimPOBackbone=Qwen2.5-7B-Instruct, Mode=SoLo2025.05 | 31 | 34.1 | 29.1 | 37.5 | 25.7 | 30.6 | |
| SFTBackbone=Llama3.1-8B-Instruct, Mode=long2025.05 | 30.9 | 36.8 | 27.3 | 36 | 29.6 | 24.8 | |
| SFTBackbone=Qwen2.5-7B-Instruct, Mode=short2025.05 | 30.8 | 33.6 | 29.1 | 39.2 | 25.7 | 27.2 | |
| SFTBackbone=Qwen2.5-7B-Instruct, Mode=long2025.05 | 30 | 32 | 28.8 | 35.7 | 25.9 | 28.7 | |
| DPOBackbone=Qwen2.5-7B-Instruct, Mode=long2025.05 | 29.7 | 34.3 | 26.9 | 35.9 | 25.6 | 27.6 | |
| LongPO(reimp)Backbone=Qwen2.5-7B-Instruct2025.05 | 29.6 | 32.2 | 28 | 36.7 | 26.7 | 23.7 | |
| InstructBackbone=Qwen2.5-7B-Instruct2025.05 | 29.3 | 30.9 | 28.3 | 36.9 | 24.6 | 26.1 | |
| ORPOBackbone=Qwen2.5-7B-Instruct, Mode=short2025.05 | 29.3 | 34.4 | 26.2 | 35 | 25.3 | 27.8 | |
| ORPOBackbone=Qwen2.5-7B-Instruct, Mode=long2025.05 | 26.6 | 30.1 | 24.5 | 33.8 | 22.3 | 23.3 | |
| DPOBackbone=Qwen2.5-7B-Instruct, Mode=short2025.05 | 25.5 | 27.2 | 24.5 | 29.7 | 23.5 | 22.8 | |
| SimPOBackbone=Qwen2.5-7B-Instruct, Mode=long2025.05 | 25.4 | 25.4 | 25.4 | 33 | — | 23.3 | |
| SimPOBackbone=Qwen2.5-7B-Instruct, Mode=short2025.05 | 24.6 | 27.1 | 23.1 | 29.5 | 21.2 | 23.1 |