Question Answering on CSQA (ACC, Length)
86.7Accuracy (CSQA)ADaPT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ADaPTBackbone=Qwen2.5-7B2026.06 | 86.7 | 65 | |
| ARMBackbone=Qwen2.5-7B2026.06 | 86.2 | 137 | |
| ADaPTanswerBackbone=Qwen2.5-7B2026.06 | 86.1 | 24 | |
| R-4BBackbone=Qwen2.5-7B2026.06 | 85.7 | 145 | |
| SFT+GRPOBackbone=Qwen2.5-7B2026.06 | 84.3 | 391 | |
| ADaPTthinkBackbone=Qwen2.5-7B2026.06 | 83.5 | 385 | |
| TLMREBackbone=Qwen2.5-7B2026.06 | 82.8 | 204 | |
| ADaPTBackbone=Qwen2.5-3B2026.06 | 79.5 | 72 | |
| SFTBackbone=Qwen2.5-7B2026.06 | 79.4 | 376 | |
| SFT+GRPOBackbone=Qwen2.5-3B2026.06 | 79.3 | 418 | |
| ARMBackbone=Qwen2.5-3B2026.06 | 78.4 | 141 | |
| ADaPTthinkBackbone=Qwen2.5-3B2026.06 | 77.9 | 364 | |
| ADaPTanswerBackbone=Qwen2.5-3B2026.06 | 77.1 | 22 | |
| R-4BBackbone=Qwen2.5-3B2026.06 | 76.9 | 226 | |
| BaseBackbone=Qwen2.5-7B2026.06 | 76.3 | 222 | |
| TLMREBackbone=Qwen2.5-3B2026.06 | 76.3 | 307 | |
| SFTBackbone=Qwen2.5-3B2026.06 | 71.4 | 164 | |
| BaseBackbone=Qwen2.5-3B2026.06 | 66.7 | 176 | |
| EMM2026.06 | 66.3 | — | |
| MERGEvolve2026.06 | 63.7 | — | |
| LoraHub2026.06 | 63.66 | — | |
| Model Swarms2026.06 | 63.66 | — | |
| Expert Fusion2026.06 | 63.22 | — | |
| Pack of LLMs2026.06 | 63.1 | — |