KG-grounded multi-hop reasoning on KG-grounded multi-hop reasoning 2-hop
61.24AccuracySPARK
Evaluation Results
| Method | Links | |
|---|---|---|
| SPARKParams=4B2026.05 | 61.24 | |
| QWEN3-VL-4B + SFT ONLYParams=4B2026.05 | 55.18 | |
| QWEN3-VL-4B (VANILLA)Params=4B2026.05 | 54.73 | |
| QWEN3-VL-4B + SPIN (NO KG)Params=4B2026.05 | 54.73 |