Multi-Route Generation on 10,000 Samples (test)
97.1Connectivity4B-Joint
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 4B-JointModel Scale=4B, Fine-tuning Data=Combined SFT data of all three tasks2026.05 | 97.1 | 98.8 | 91.7 | 79.4 | 80.8 | 67.2 | 98.1 | 1.33 | 54.7 | |
| Qwen3-4BModel Scale=4B, CPT Data Fraction=100%2026.05 | 96.3 | 98 | 90.1 | 78.2 | 79.4 | 64.5 | 98 | 1.45 | 54.5 | |
| Qwen3-1.7BModel Scale=1.7B2026.05 | 95.2 | 95.6 | 80.6 | 75.7 | 75 | 52.9 | 97.6 | 1.74 | 53.2 | |
| Qwen3-4B-25Model Scale=4B, CPT Data Fraction=25%2026.05 | 94.9 | 97 | 84.8 | 75.6 | 76.1 | 57.9 | 97.4 | 1.9 | 53.4 | |
| Qwen3-0.6BModel Scale=0.6B2026.05 | 92.3 | 95.4 | 81 | 75.3 | 74.7 | 52.6 | 96.3 | 2.28 | 51.4 |