Reranking on Amazon Beauty (test)
80.3NDCGGemma-4
Evaluation Results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Gemma-4Model size=26B-E4B, Training=—2026.07 | 80.3 | 78.27 | 67.81 | 65.6 | 89.29 | 96.89 | 21.87 | 55.27 | 73.66 | 65.6 | 55.27 | 44.19 | 0 | 10 | |
| SCOReD (Qwen3-0.6B)Model size=0.6B, Training=SFT2026.07 | 79.08 | 77.25 | 66.67 | 64.53 | 88.65 | 95.64 | 21.51 | 54.45 | 72.43 | 64.53 | 54.45 | 43.46 | 1.52 | 6.2 | |
| Qwen-3.6Model size=35B-A3B, Training=—2026.07 | 78.79 | 76.83 | 66.66 | 64.55 | 87.35 | 94.64 | 21.52 | 54.43 | 72.21 | 64.55 | 54.43 | 43.33 | 2.1 | 9.4 | |
| Baseline (Qwen3-0.6B)Model size=0.6B, Training=SFT2026.07 | 77.86 | 75.92 | 65.58 | 63.41 | 86.57 | 94.35 | 21.14 | 53.35 | 71.08 | 63.41 | 53.35 | 42.65 | 2.82 | 8.5 | |
| LLM-Summ. (Qwen3-0.6B)Model size=0.6B, Training=SFT2026.07 | 73.42 | 71.79 | 61.97 | 60.21 | 82.03 | 89.2 | 20.07 | 50.63 | 67.46 | 60.21 | 50.63 | 40.47 | 8.74 | 4.4 |