Social Interaction Question Answering on SocialIQA (test)
75.49AccuracyTaT
Evaluation results
| Method | Links | |
|---|---|---|
| TaTTrain Dataset=ARC-C2026.03 | 75.49 | |
| Expert III (SocialIQA)Seed Model=LLaMA-3.2-3B2026.05 | 72.42 | |
| MetaMoESeed Model=LLaMA-3.2-3B2026.05 | 72.26 | |
| UnrestrictedMoESeed Model=LLaMA-3.2-3B2026.05 | 71.8 | |
| ModelSoupSeed Model=LLaMA-3.2-3B2026.05 | 71.75 | |
| BTXSeed Model=LLaMA-3.2-3B2026.05 | 71.75 | |
| FlexOlmoSeed Model=LLaMA-3.2-3B2026.05 | 70.88 | |
| TaTTrain Dataset=OpenQA2026.03 | 69.7 | |
| Linear ProbeTrain Dataset=SiQA2026.03 | 69.33 | |
| BTMSeed Model=LLaMA-3.2-3B2026.05 | 68.17 | |
| Few-shot AccuracyMode=Few-shot2026.03 | 67.7 | |
| Linear ProbeTrain Dataset=CosQA2026.03 | 67.69 | |
| Linear ProbeTrain Dataset=ARC-E2026.03 | 66.6 | |
| Linear ProbeTrain Dataset=ComQA2026.03 | 66.38 | |
| TaTTrain Dataset=SiQA2026.03 | 66.17 | |
| Linear ProbeTrain Dataset=ARC-C2026.03 | 65.47 | |
| Linear ProbeTrain Dataset=OpenQA2026.03 | 65.47 | |
| Zero-shot AccuracyMode=Zero-shot2026.03 | 65.2 | |
| TaTTrain Dataset=CosQA2026.03 | 61.21 | |
| TaTTrain Dataset=ARC-E2026.03 | 60.85 | |
| Expert I (CSQA)Seed Model=LLaMA-3.2-3B2026.05 | 60.54 | |
| Linear ProbeTrain Dataset=Hellaswag2026.03 | 60.39 | |
| TaTTrain Dataset=ComQA2026.03 | 60.18 | |
| Expert II (CosmosQA)Seed Model=LLaMA-3.2-3B2026.05 | 58.96 | |
| ZeroShotSeed Model=LLaMA-3.2-3B2026.05 | 56.19 | |
| TaTTrain Dataset=Hellaswag2026.03 | 55.27 | |
| TaTTrain Dataset=BoolQ2026.03 | 50.2 | |
| Linear ProbeTrain Dataset=BoolQ2026.03 | 49.2 |