Question Answering on Pop-QA Cities-20
24.2BLEU-1Knowledge-Instruct
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Knowledge-InstructSize=5M2026.05 | 24.2 | 14.1 | 6.4 | 32 | 13.3 | 25.7 | 30.7 | |
| Knowledge-InstructSize=20M2026.05 | 23.9 | 14.1 | 6.5 | 31 | 12.9 | 24.6 | 30.8 | |
| InstructLabSize=5M2026.05 | 22.1 | 14.4 | 7.7 | 36.3 | 17.2 | 30.9 | 25.8 | |
| InstructLabSize=20M2026.05 | 21.3 | 14.1 | 7.3 | 35.6 | 17.3 | 30.5 | 26.1 | |
| EmbGenSize=20M2026.05 | 20.1 | 12.2 | 5.9 | 33.8 | 14.4 | 28.4 | 22.8 | |
| EmbGenSize=5M2026.05 | 16.5 | 9.3 | 4.4 | 29.2 | 10.6 | 24.2 | 21.8 | |
| EntiGraphSize=5M2026.05 | 13.5 | 7.6 | 2.8 | 20.3 | 7.6 | 15.3 | 23.7 | |
| EntiGraphSize=20M2026.05 | 12.8 | 7.1 | 2.8 | 19.5 | 7.1 | 14.8 | 22.4 | |
| Baseline: llama 8B InstructSize=5M2026.05 | 10.8 | 6.3 | 2.6 | 18.2 | 6.8 | 14.6 | 25.1 | |
| Baseline: llama 8B InstructSize=20M2026.05 | 10.8 | 6.3 | 2.6 | 18.2 | 6.8 | 14.6 | 25.1 |