RTL code functionality equivalence checking on DeepRTL2 Benchmark
74.5APUniRTL
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| UniRTLDesign Format=code & graph2026.05 | 74.5 | 74.7 | 75.3 | 73.4 | 77.3 | |
| GraphCodeBERTDesign Format=code & graph2026.05 | 73 | 73.3 | 75.3 | 61.3 | 97.3 | |
| UniRTL (w/o graph)Design Format=code2026.05 | 71.2 | 66.7 | 71.7 | 57.7 | 94.7 | |
| DeepRTL2backbone=Llama2025.05 | 66.7 | — | — | — | — | |
| DeepRTL2-LlamaDesign Format=code2026.05 | 64.6 | 69.5 | 73.7 | 59.7 | 96.4 | |
| DeepRTL2-DeepSeekDesign Format=code2026.05 | 63.1 | 64 | 72.9 | 58.7 | 96 | |
| CircuitFusionDesign Format=code & graph2026.05 | 62.8 | 66.7 | 73.6 | 60.2 | 94.7 | |
| GritLM-7BDesign Format=code2026.05 | 59.9 | 64 | 72.4 | 58.7 | 94.7 | |
| DeepRTL2backbone=DeepSeek2025.05 | 59.1 | — | — | — | — | |
| text-embedding-3-smallsize=small2025.05 | 56.5 | — | — | — | — | |
| text-embedding-3-largeDesign Format=code2026.05 | 56.4 | 58.7 | 68.7 | 55.3 | 90.7 | |
| NV-Embed-v2Design Format=code2026.05 | 55.4 | 60.7 | 66.7 | 54.7 | 85.3 | |
| text-embedding-3-smallDesign Format=code2026.05 | 54.3 | 61.3 | 69.6 | 54.5 | 96 | |
| GritLMparameters=7B2025.05 | 54.1 | — | — | — | — | |
| DeepRTL2backbone=Llama, hard negatives=false2025.05 | 51.8 | — | — | — | — | |
| text-embedding-3-largesize=large2025.05 | 49.8 | — | — | — | — | |
| DeepRTL2backbone=DeepSeek, hard negatives=false2025.05 | 48.1 | — | — | — | — |