Static model relationship recognition on Bench-A 20 fixed (train test)
85AccuracyPhyloLM
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PhyloLM2026.06 | 85 | 89.8 | 79.2 | 83.4 | 94.9 | |
| MPT2026.06 | 84.6 | 89.7 | 78.3 | 82.8 | 94 | |
| Per-sample proxy comparisonProxy=Qwen3-8B, M=42026.06 | 79.6 | 76.3 | 90.8 | 82 | 89.6 | |
| Per-sample proxy comparisonProxy=Qwen3.5-9B, M=42026.06 | 77.9 | 83.6 | 71.7 | 75.8 | 89 | |
| LLM-DNA-splitEncoder=MPNet, Aggregation=prompt-aligned variant2026.06 | 75.4 | 87.8 | 59.2 | 68.6 | 81.4 | |
| LLM-DNAEncoder=Qwen3-Emb-8B, Aggregation=original model-level RGP2026.06 | 75 | 80.1 | 67.5 | 72.1 | 79 | |
| LLM-DNA-splitEncoder=Qwen3-Emb-8B, Aggregation=prompt-aligned variant2026.06 | 74.6 | 85.3 | 59.2 | 67.9 | 80.3 | |
| LLM-DNA-splitEncoder=BGE, Aggregation=prompt-aligned variant2026.06 | 74.2 | 88.7 | 55.8 | 66.4 | 83.1 | |
| LLM-DNAEncoder=BGE, Aggregation=original model-level RGP2026.06 | 72.5 | 78.7 | 68.3 | 69.6 | 85.3 | |
| LLM-DNAEncoder=MPNet, Aggregation=original model-level RGP2026.06 | 67.1 | 80 | 48.3 | 59 | 77.2 |