Open-Ended Abstract Retrieval (OAR) on Music
49.2DiversityR4T-FOLM
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| R4T-FOLMBackbone=Gemma3-4B2026.03 | 49.2 | 63.1 | 62 | 58.1 | |
| Gemini-2.5-FlashStrategy=Best-of-N2026.03 | 48.4 | 48.2 | 49 | 48.5 | |
| R4T-DiffusionBackbone=Gemma3-4B2026.03 | 46.7 | — | 59.6 | — | |
| Gemini-2.5-FlashStrategy=Zero-shot2026.03 | 45.2 | 45.8 | 44.4 | 45.1 | |
| R4T-FOLMBackbone=Qwen3-4B2026.03 | 44.8 | 48.2 | 49.4 | 47.5 | |
| R4T-DiffusionBackbone=Qwen3-4B2026.03 | 44.5 | — | 52 | — | |
| Gemma3-4BStrategy=Best-of-N2026.03 | 43.2 | 51.4 | 53 | 49.2 | |
| Gemma3-4BStrategy=Zero-shot2026.03 | 42.6 | 49.8 | 51.8 | 48.1 | |
| Qwen3-4BStrategy=Best-of-N2026.03 | 40.3 | 44 | 43.7 | 42.7 | |
| Qwen3-4BStrategy=Zero-shot2026.03 | 38.8 | 42 | 41.2 | 40.7 | |
| No Fan-out2026.03 | 20 | 48.8 | 41.8 | 36.9 |