Multi-hop Question Answering on MuSiQue (Token-level F1)
38.1Token-level F1PDA-GAM
Evaluation Results
| Method | Links | |
|---|---|---|
| PDA-GAMStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=42026.05 | 38.1 | |
| PDA-GAMStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=32026.05 | 37.9 | |
| PDA-TIESStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=32026.05 | 37.8 | |
| PDA-TIESStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=42026.05 | 37.8 | |
| PDA-KnOTSStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=42026.05 | 37 | |
| PDA-AveragingStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=42026.05 | 36.9 | |
| PDA-GAMStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=22026.05 | 36.1 | |
| PDA-TIESStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=22026.05 | 35.6 | |
| Search-R1Student Model=Qwen3-8B, Preference Delta Scale=7B over 3B, Number of Preference Deltas (∆)=12026.05 | 34.7 | |
| Qwen3Student Model=Qwen3-8B, Preference Delta Scale=4B over 1.7B, Number of Preference Deltas (∆)=12026.05 | 32.8 | |
| PDA-GAMStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=42026.05 | 32.8 | |
| PDA-GAMStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=32026.05 | 32.4 | |
| PDA-TIESStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=42026.05 | 32.1 | |
| Joint ORPOStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=42026.05 | 32 | |
| PDA-AveragingStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=42026.05 | 32 | |
| PDA-TIESStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=32026.05 | 32 | |
| PDA-KnOTSStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=42026.05 | 31.9 | |
| PDA-TIESStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=22026.05 | 31.6 | |
| AceSearcherStudent Model=Qwen3-8B, Preference Delta Scale=7B over 3B, Number of Preference Deltas (∆)=12026.05 | 30.5 | |
| PDA-GAMStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=22026.05 | 30.3 | |
| Search-R1Student Model=Tülu3-8B, Preference Delta Scale=7B over 3B, Number of Preference Deltas (∆)=12026.05 | 29.5 | |
| Llama-3.2Student Model=Qwen3-8B, Preference Delta Scale=3B over 1B, Number of Preference Deltas (∆)=12026.05 | 29.1 | |
| Student BaselineStudent Model=Qwen3-8B2026.05 | 28.9 | |
| Joint ORPOStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=42026.05 | 28.9 | |
| Qwen3Student Model=Tülu3-8B, Preference Delta Scale=4B over 1.7B, Number of Preference Deltas (∆)=12026.05 | 28.5 | |
| AceSearcherStudent Model=Tülu3-8B, Preference Delta Scale=7B over 3B, Number of Preference Deltas (∆)=12026.05 | 27.8 | |
| Sequential ORPOStudent Model=Qwen3-8B, Number of Preference Deltas (∆)=22026.05 | 27.6 | |
| Llama-3.2Student Model=Tülu3-8B, Preference Delta Scale=3B over 1B, Number of Preference Deltas (∆)=12026.05 | 26.2 | |
| Student BaselineStudent Model=Tülu3-8B2026.05 | 25.6 | |
| Sequential ORPOStudent Model=Tülu3-8B, Number of Preference Deltas (∆)=22026.05 | 21.4 |