Code-to-Text generation on CodeXGLUE (test)
20.57PythonCodeT5
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CodeT5model_size=large2022.07 | 20.57 | 15.58 | 16.17 | 19.69 | 20.74 | 26.49 | 19.87 | — | — | |
| CodeT5model_size=small2022.07 | 20.04 | 14.87 | 15.32 | 19.25 | 19.92 | 25.46 | 19.14 | — | — | |
| CodeT5model_size=base2022.07 | 20.01 | 15.24 | 16.16 | 19.56 | 20.31 | 26.03 | 19.55 | — | — | |
| CoTexT2022.07 | 19.73 | 14.02 | 14.96 | 18.86 | 19.06 | 24.58 | 18.55 | — | — | |
| PLBART2022.07 | 19.3 | 14.11 | 15.56 | 18.91 | 18.45 | 23.58 | 18.32 | — | — | |
| CodeBERT2022.07 | 19.06 | 12.16 | 14.9 | 18.07 | 17.65 | 25.16 | 17.83 | — | — | |
| DOBF2022.07 | 18.24 | — | — | — | 19.05 | — | — | — | — | |
| RoBERTa2022.07 | 18.14 | 11.17 | 11.9 | 17.72 | 16.47 | 24.02 | 16.57 | — | — | |
| R&QModel architecture=Switch Transformer (8 experts), Strategy=Replicate-and-Quantize2026.02 | — | — | — | — | — | — | — | 1.5517 | 1.9426 | |
| RawModel architecture=Switch Transformer (8 experts), Strategy=Unmodified model2026.02 | — | — | — | — | — | — | — | 1.5463 | 2.6119 |