Agentic Planning on DeepPlanning (remaining 20% tasks)
13.6Avg@16OPD + Q3-only
Evaluation Results
| Method | Links | |
|---|---|---|
| OPD + Q3-onlyStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Thinking-enabled=true, Token Retention Rate=20%, Selection Strategy=Q3 quadrant (TIP)2026.04 | 13.6 | |
| Entropy-onlyStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Thinking-enabled=true, Token Retention Rate=50%, Selection Strategy=Entropy-based2026.04 | 13.1 | |
| OPD + Soft-ORStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Thinking-enabled=true, Token Retention Rate=50%, Selection Strategy=Soft-OR2026.04 | 13.1 | |
| OPD (all tokens)Student Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Thinking-enabled=true, Token Retention Rate=100%, Selection Strategy=Full tokens2026.04 | 12.8 | |
| Entropy-onlyStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Thinking-enabled=true, Token Retention Rate=20%, Selection Strategy=Entropy-based2026.04 | 12.7 | |
| OPD + Q3-onlyStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-14B, Thinking-enabled=true, Token Retention Rate=20%, Selection Strategy=Q3 quadrant (TIP)2026.04 | 12.6 | |
| OPD + Soft-ORStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Thinking-enabled=true, Token Retention Rate=20%, Selection Strategy=Soft-OR2026.04 | 12.6 | |
| Entropy-onlyStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-14B, Thinking-enabled=true, Token Retention Rate=50%, Selection Strategy=Entropy-based2026.04 | 12.1 | |
| OPD + Soft-ORStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-14B, Thinking-enabled=true, Token Retention Rate=20%, Selection Strategy=Soft-OR2026.04 | 12.1 | |
| OPD + Soft-ORStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-14B, Thinking-enabled=true, Token Retention Rate=50%, Selection Strategy=Soft-OR2026.04 | 12 | |
| OPD (all tokens)Student Model=Qwen3-1.7B, Teacher Model=Qwen3-14B, Thinking-enabled=true, Token Retention Rate=100%, Selection Strategy=Full tokens2026.04 | 11.7 | |
| Entropy-onlyStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-14B, Thinking-enabled=true, Token Retention Rate=20%, Selection Strategy=Entropy-based2026.04 | 11.6 |