Post-synthesis Delay Prediction on RTL Design Dataset (test)
0.534MAECodeV-QW-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CodeV-QW-7BKnowledge Distillation=w/o KD, Category=LLM-Based Baselines, Backbone=CodeQwen2025.08 | 0.534 | 9 | 0.5277 | 0.6872 | |
| StructRTLKnowledge Distillation=w/o KD, Category=Ours2025.08 | 0.5414 | 10 | 0.763 | 0.4868 | |
| CodeV-CL-7BKnowledge Distillation=w/o KD, Category=LLM-Based Baselines, Backbone=CodeLlama2025.08 | 0.562 | 10 | 0.5174 | 0.6947 | |
| StructRTL (w/o Lmnm)Knowledge Distillation=w/o KD, Category=Ours, Ablation=without structural graph alignment loss2025.08 | 0.573 | 11 | 0.7473 | 0.5027 | |
| StructRTL (w/o Lep)Knowledge Distillation=w/o KD, Category=Ours, Ablation=without positional embedding loss2025.08 | 0.5902 | 11 | 0.7368 | 0.513 | |
| Graph-XGBoostKnowledge Distillation=w/o KD, Category=Graph-Based Baselines2025.08 | 0.6384 | 12 | 0.3362 | 0.8147 | |
| CodeV-DS-6.7BKnowledge Distillation=w/o KD, Category=LLM-Based Baselines, Backbone=DeepSeek-Coder2025.08 | 0.6403 | 12 | 0.3905 | 0.7807 | |
| Graph-GNNKnowledge Distillation=w/o KD, Category=Graph-Based Baselines2025.08 | 0.7327 | 13 | 0.6639 | 0.5797 |