KG-grounded multi-hop reasoning
66.89FaithfulnessSPARK
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SPARKParams=4B2026.05 | 66.89 | 9.09 | |
| QWEN3-VL-4B + SFT ONLYParams=4B2026.05 | 60.5 | 15.97 | |
| QWEN3-VL-4B + SPIN (NO KG)Params=4B2026.05 | 60.43 | 12.7 | |
| QWEN3-VL-4B (VANILLA)Params=4B2026.05 | 59.45 | 12.7 |