Multilingual Question Answering on MKQA
57.82Composite Score (ALL-AVG)TR-RAG
Evaluation Results
| Method | Links | |||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TR-RAGBackbone=Llama-3.2-3B-Instruct, Evidence Language=English2026.07 | 57.82 | 50.11 | 57.62 | 58.67 | 55.47 | 69.57 | 55.95 | 55 | 60.17 | 60.05 | 74.4 | 77.05 | 70.5 | 72.8 | 69.63 | 77.93 | 73.45 | 71.98 | 58.1 | 74.6 | 79.15 | 70.62 | 75.2 | 68.05 | 75.45 | 72.9 | 71.76 | — | — | |
| TR-RAGBackbone=Llama-3.2-3B-Instruct2026.07 | 57.82 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 71.98 | — | — | — | — | — | — | — | — | 71.76 | 44.5 | 43.05 | |
| Naive-RLBackbone=Llama-3.2-3B-Instruct, Evidence Language=English2026.07 | 57.18 | 50.8 | 55.83 | 58 | 54.87 | 70.27 | 55.54 | 52.62 | 59.48 | 62.35 | 71.45 | 78.5 | 70.77 | 73.1 | 67.83 | 74.38 | 71.77 | 71.27 | 57.9 | 71.25 | 77.95 | 69.03 | 76.9 | 67.35 | 72.15 | 72.13 | 70.58 | — | — | |
| Naive-RLBackbone=Llama-3.2-3B-Instruct2026.07 | 57.18 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 71.27 | — | — | — | — | — | — | — | — | 70.58 | 44 | 42.85 | |
| BaseBackbone=Llama-3.2-3B-Instruct, Evidence Language=English2026.07 | 53.7 | 49.72 | 50.62 | 53.55 | 51.3 | 67.35 | 53.62 | 47.36 | 56.11 | 58.35 | 63.58 | 73.13 | 65.02 | 68.96 | 62.93 | 67.16 | 66.35 | 65.69 | 52.5 | 62.85 | 71.77 | 62.37 | 70.86 | 60.7 | 63.8 | 65.12 | 63.75 | — | — | |
| BaseBackbone=Llama-3.2-3B-Instruct2026.07 | 53.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 65.69 | — | — | — | — | — | — | — | — | 63.75 | 44 | 41.39 |