Natural Language Understanding on GLUE (RTE, SST2, MRPC, STS-B Detailed Metrics)
65.3RTE AccuracyGiLT-GPT2
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GiLT-GPT2Base Model=GPT2-medium (355M), Architecture=replaced last 12 vanilla transformer layers with Graph-Infused layers, Training=separately finetuned on each downstream task2026.05 | 65.3 | 95.11 | 81.39 | 86.97 | 85.2 | 84.3 | |
| Post-GPT2Base Model=GPT2-medium (355M), Training=separately finetuned on each downstream task2026.05 | 64.1 | 94.84 | 80.75 | 86.29 | 84.2 | 83.6 |