Structured Flow Evaluation on StructFlowBench
94.25CSRGraphIF
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GraphIFBackbone Model=Llama-3.1-70B-Instruct2025.11 | 94.25 | 81.39 | 94.1 | 94.25 | |
| GraphIFBackbone Model=GPT-4o-mini2025.11 | 91.16 | 77.6 | 93.38 | 91.16 | |
| Qwen2.5-7B+GraphIFBackbone=Qwen2.5-7B, Technique=GraphIF2025.11 | 89.46 | 69.25 | 89.47 | 88.6 | |
| LLM-onlyBackbone Model=Llama-3.1-70B-Instruct2025.11 | 87.18 | 64.72 | 87.74 | 87.18 | |
| LLM-onlyBackbone Model=GPT-4o-mini2025.11 | 79.19 | 41.93 | 81.17 | 79.19 | |
| MemAgent-7BBackbone=7B2025.11 | 75.56 | 32.81 | 76.06 | 75.56 | |
| Qwen2.5-7B+Self-RefineBackbone=Qwen2.5-7B, Technique=Self-Refine2025.11 | 74.75 | 29.95 | 76.28 | 74.75 | |
| Qwen2.5-7B-onlyBackbone=Qwen2.5-7B2025.11 | 70.7 | 23.02 | 71.22 | 67.07 |