Agentic Workflow Performance Prediction on FLORA-Bench Math-AF 1.0 (test)
81.3AccuracyGLOW
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GLOWbase_LLM=Qwen3-1.7B, GNN=two-layer GAT, SBERT=all-MiniLM-L6-v22025.12 | 81.3 | 75.1 | |
| GCN2025.12 | 79.8 | 72.9 | |
| Agentic Predictor2025.12 | 79.8 | 73.4 | |
| GAT2025.12 | 79.4 | 72.2 | |
| Graph Transformer2025.12 | 79.4 | 71.4 | |
| One-For-All2025.12 | 78.9 | 69.8 | |
| GCNII2025.12 | 78.4 | 72.5 | |
| Qwen3base_model=Qwen3-1.7B, fine-tuned=true2025.12 | 76.7 | 70.4 |