KG-grounded Multi-hop Reasoning on KG-grounded multi-hop reasoning 1-hop
63.67AccuracySPARK
Evaluation Results
| Method | Links | |
|---|---|---|
| SPARKParams=4B2026.05 | 63.67 | |
| QWEN3-VL-4B + SFT ONLYParams=4B2026.05 | 59.9 | |
| QWEN3-VL-4B (VANILLA)Params=4B2026.05 | 59.33 | |
| QWEN3-VL-4B + SPIN (NO KG)Params=4B2026.05 | 57.27 |