Multilingual Question Answering on TyDiQA GoldP (val)
80Ar ScoreByT5 XXL
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| ByT5 XXLEvaluation Protocol=finetuned2022.04 | 80 | 85 | 77.7 | 78.8 | 85.7 | 78.3 | 78.2 | 84 | 85.5 | 81.4 | |
| mT5 XXLEvaluation Protocol=finetuned2022.04 | 76.9 | 80.5 | 75.5 | 76.3 | 81.8 | 75.7 | 76.8 | 84.4 | 83.9 | 79.1 | |
| PaLM 540BEvaluation Protocol=finetuned2022.04 | 75 | 83.2 | 75.5 | 78.9 | 84.1 | 75.7 | 77.1 | 85.2 | 84.9 | 80 | |
| PaLM 540BEvaluation Protocol=few-shot2022.04 | 56.4 | 54 | 65.5 | 66.4 | 69.2 | 63.8 | 46.8 | 75.6 | 46.9 | 60.5 | |
| RISEBackbone=Qwen3-30B-A3B, Budget=128, Target Language=ID2026.04 | 51.59 | 53.97 | 25.65 | 23.58 | 31.79 | 47.75 | 29.54 | 30.13 | 59.95 | 39.33 | |
| RISEBackbone=Qwen3-30B-A3B, Budget=128, Target Language=RU2026.04 | 51.06 | 53.62 | 23.88 | 24.15 | 29.55 | 49.08 | 29.62 | 29.3 | 60.66 | 38.99 | |
| RandomBackbone=Qwen3-30B-A3B, Budget=128, Target Language=BN2026.04 | 50.31 | 53.52 | 24.43 | 22.69 | 26.49 | 48.31 | 26.08 | 27.22 | 59.71 | 37.64 | |
| TopKBackbone=Qwen3-30B-A3B, Budget=128, Target Language=BN2026.04 | 50.2 | 52.74 | 24.75 | 23.13 | 27.86 | 47.51 | 26.94 | 26.85 | 59.11 | 37.68 | |
| ESFTBackbone=Qwen3-30B-A3B2026.04 | 49.51 | 51.79 | 24.08 | 23.26 | 26.89 | 47.33 | 27.09 | 25.85 | 59.7 | 37.28 | |
| VanillaBackbone=Qwen3-30B-A3B2026.04 | 49.49 | 51.51 | 24.15 | 22.98 | 26.86 | 48.35 | 27.04 | 27.05 | 58.61 | 37.34 | |
| RISEBackbone=Qwen3-30B-A3B, Budget=128, Target Language=BN2026.04 | 49.15 | 54.23 | 24.04 | 23.39 | 27.61 | 47.77 | 27.22 | 27.4 | 60.39 | 37.91 | |
| RandomBackbone=Qwen3-30B-A3B, Budget=64, Target Language=BN2026.04 | 48.91 | 52.17 | 23.87 | 22.31 | 25.61 | 48.19 | 25.63 | 27.73 | 59.03 | 37.05 | |
| TopKBackbone=Phi-3.5-MoE-Instruct, Budget=16, Target Language=BN2026.04 | 39.63 | 49.51 | 18.32 | 20.78 | 23.59 | 34.58 | 18.91 | 11.56 | 10.78 | 25.3 | |
| RISEBackbone=Phi-3.5-MoE-Instruct, Budget=16, Target Language=ID2026.04 | 39.43 | 36.83 | 19 | 21.51 | 22.33 | 24.5 | 20.78 | 11.03 | 8.49 | 22.66 | |
| ESFTBackbone=Phi-3.5-MoE-Instruct2026.04 | 38.29 | 45.44 | 18.1 | 20.12 | 23.15 | 28.84 | 19.1 | 12.24 | 10.09 | 23.93 | |
| RandomBackbone=Phi-3.5-MoE-Instruct, Budget=16, Target Language=BN2026.04 | 38.26 | 41.93 | 18.9 | 19.9 | 23.23 | 26.84 | 18.75 | 11.64 | 9.34 | 23.2 | |
| RISEBackbone=Phi-3.5-MoE-Instruct, Budget=16, Target Language=RU2026.04 | 37.68 | 34.83 | 17.63 | 20.85 | 22.85 | 21.98 | 17.81 | 10.7 | 8.32 | 21.41 | |
| RandomBackbone=Phi-3.5-MoE-Instruct, Budget=32, Target Language=BN2026.04 | 37.53 | 43.61 | 18.89 | 20.1 | 23.28 | 24.56 | 18.45 | 11.44 | 8.93 | 22.98 | |
| RISEBackbone=Phi-3.5-MoE-Instruct, Budget=16, Target Language=BN2026.04 | 37.43 | 46.89 | 17.55 | 19.64 | 22.75 | 21.1 | 17.26 | 11.56 | 9.68 | 22.65 | |
| VanillaBackbone=Phi-3.5-MoE-Instruct2026.04 | 36.75 | 36.04 | 16.94 | 19.19 | 22.09 | 19.75 | 16.19 | 11.55 | 7.68 | 20.69 |