Commonsense Reasoning on X-CSQA (test)
62.5Average AccuracyST-Merge
Evaluation Results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ST-Merge2026.06 | 62.5 | 53.6 | — | — | — | — | — | — | — | — | — | — | — | — | 67.2 | — | 77.3 | |
| LayAlign2026.06 | 62.3 | 53.3 | — | — | — | — | — | — | — | — | — | — | — | — | 66.5 | — | 76.7 | |
| MindMerger-Softtype=Soft interpolation2024.05 | 61 | 45.5 | 46.2 | 48.4 | 51.4 | 60.6 | 53.9 | 63.3 | 62.9 | 63.8 | 63.7 | 66.8 | 67 | 67.1 | 68.1 | 69.1 | 78.1 | |
| MindMerger2026.06 | 61 | 45.5 | — | — | — | — | — | — | — | — | — | — | — | — | 68.1 | — | 78.1 | |
| polyGRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 60.46 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 57.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MultiReason-Loratuning=LoRA2024.05 | 56.9 | 25.1 | 32 | 39.2 | 42.2 | 56.6 | 55.9 | 60.6 | 62.2 | 61.3 | 62.8 | 66.3 | 64.9 | 66.2 | 67.4 | 67.7 | 79.3 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.04 | 54.27 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| polyGRPOBackbone=Llama3-8B-Instruct2026.04 | 53.64 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GRPOBackbone=Llama3-8B-Instruct2026.04 | 53.36 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LIDRBackbone=Qwen2.5-7B-Instruct2026.04 | 53.18 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MindMerger-Hardtype=Hard interpolation2024.05 | 53.1 | 33.1 | 29.9 | 40.4 | 37.7 | 52.9 | 49.9 | 54.7 | 55.4 | 58 | 58 | 59.7 | 58.6 | 61.9 | 62.5 | 63.6 | 75.2 | |
| QAlign2024.05 | 52.3 | 35.1 | 32.6 | 37.8 | 36.3 | 50.5 | 49.2 | 51.3 | 54.8 | 56.3 | 56.3 | 58.3 | 58.8 | 59.8 | 60.3 | 63.1 | 75.7 | |
| Translate-Enstrategy=Translation to English2024.05 | 52.3 | 36.5 | 41.3 | 48.4 | 44.6 | 51.8 | 47.1 | 53.3 | 51.5 | 55 | 54.4 | 56.3 | 57.3 | 54.7 | 57.2 | 55.5 | 71.3 | |
| Translate-En2026.06 | 52.3 | 36.5 | — | — | — | — | — | — | — | — | — | — | — | — | 57.2 | — | 71.3 | |
| QAlign2026.06 | 52.3 | 35.1 | — | — | — | — | — | — | — | — | — | — | — | — | 60.3 | — | 75.7 | |
| LIDRBackbone=Llama3-8B-Instruct2026.04 | 52.22 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MonoReason2024.05 | 51.3 | 24.2 | 25.1 | 32.9 | 32.3 | 50.9 | 49.1 | 50.6 | 56.5 | 57.5 | 56 | 56 | 61.2 | 61.7 | 63.5 | 64 | 76.3 | |
| MetaMath2026.06 | 51.3 | 24.2 | — | — | — | — | — | — | — | — | — | — | — | — | 63.5 | — | 76.3 | |
| MAPOBackbone=Qwen2.5-7B-Instruct2026.04 | 50.69 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| xRFTBackbone=Qwen2.5-7B-Instruct2026.04 | 49.25 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| xRFTBackbone=Llama3-8B-Instruct2026.04 | 48.15 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BaseBackbone=Llama3-8B-Instruct2026.04 | 45.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAPOBackbone=Llama3-8B-Instruct2026.04 | 43.89 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MultiReason-SFTtuning=SFT2024.05 | 43.8 | 27.6 | 29.2 | 32 | 28.7 | 38.8 | 38.7 | 45.5 | 43.8 | 45.9 | 46.5 | 50.2 | 49.1 | 51.2 | 52.1 | 54.3 | 67.2 | |
| MultiReason2026.06 | 43.8 | 27.6 | — | — | — | — | — | — | — | — | — | — | — | — | 52.1 | — | 67.2 | |
| LangBridge2024.05 | 36.1 | 31.8 | 30.5 | 30.6 | 30.6 | 33.3 | 33.9 | 39.8 | 39.8 | 38.4 | 35.1 | 39.1 | 37.4 | 36.3 | 38.2 | 38.4 | 44.4 | |
| LangBridge2026.06 | 36.1 | 31.8 | — | — | — | — | — | — | — | — | — | — | — | — | 38.2 | — | 44.4 |