Multi-turn Code-Switched Question Answering on PingPong answerable QA (Hausa–English ha-en) (test)
90.4AccuracyQwen3.5-27B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3.5-27BSetting=BASE2026.06 | 90.4 | |
| Qwen3.5-27BSetting=CANVAS2026.06 | 90.4 | |
| Llama-3.1-8BSetting=CANVAS2026.06 | 65.66 | |
| Llama-3.1-8BSetting=BASE2026.06 | 60.61 | |
| Aya-Expanse-8BSetting=CANVAS2026.06 | 50 | |
| Aya-Expanse-8BSetting=BASE2026.06 | 44.44 | |
| Mistral-7BSetting=CANVAS2026.06 | 40.4 | |
| Mistral-7BSetting=BASE2026.06 | 39.9 |