Knowledge-intensive Question Answering on GPQA (Accuracy)
48.5AccuracyDIVER
Evaluation Results
| Method | Links | |
|---|---|---|
| DIVERBackbone=Qwen3-4B2026.05 | 48.5 | |
| GCPOBackbone=Qwen3-4B2026.05 | 47.5 | |
| DQOBackbone=Qwen3-4B2026.05 | 47.2 | |
| DAPOBackbone=Qwen3-4B2026.05 | 46 | |
| Div-R1Backbone=Qwen3-4B2026.05 | 45.2 | |
| GRPOBackbone=Qwen3-4B2026.05 | 44.4 | |
| GCPOBackbone=Qwen3-1.7B2026.05 | 31.8 | |
| DIVERBackbone=Qwen3-1.7B2026.05 | 30.2 | |
| DQOBackbone=Qwen3-1.7B2026.05 | 28.3 | |
| DAPOBackbone=Qwen3-1.7B2026.05 | 28.2 | |
| Qwen3-4B(Base)Backbone=Qwen3-4B2026.05 | 26.3 | |
| Div-R1Backbone=Qwen3-1.7B2026.05 | 25.1 | |
| GRPOBackbone=Qwen3-1.7B2026.05 | 24.3 | |
| Qwen3-1.7B(Base)Backbone=Qwen3-1.7B2026.05 | 20.2 |