Algorithmic Reasoning on CLRS
99.8BFS Success RateMPNN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MPNNModel Type=GNN2025.11 | 99.8 | 99.4 | 99.2 | 99.6 | |
| AutoBRANE (Algorithm 3)Base Model=Llama-3-8B, fine-tuning=LoRA + GNN embeddings2025.11 | 99.3 | 92.6 | 78.7 | 79.4 | |
| LoRABase Model=Llama-3-8B, fine-tuning=parameter-efficient2025.11 | 98.2 | 89.2 | 75.6 | 78.5 | |
| AdapterBase Model=Llama-3-8B, fine-tuning=parameter-efficient2025.11 | 96.5 | 83.1 | 69.4 | 74.3 | |
| PrefixTuningBase Model=Llama-3-8B, fine-tuning=parameter-efficient2025.11 | 92.1 | 80.2 | 65.3 | 74.6 | |
| PromptingBase Model=Llama-3-8B, m (examples)=202025.11 | 35.88 | 33.56 | 25.8 | 26.59 | |
| PromptingBase Model=Llama-3-8B, m (examples)=102025.11 | 34.65 | 32.44 | 25.32 | 25.98 | |
| PromptingBase Model=Llama-3-8B, m (examples)=52025.11 | 34.28 | 31.54 | 25.17 | 25.31 | |
| PromptingBase Model=Llama-3-8B, m (examples)=02025.11 | 22.47 | 2.78 | 4.6 | 3.95 |