Natural Language Understanding on GLUE v1 (test dev)
96.22SST-2 AccuracyLoRA
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LoRATuning=PEFT, Tunable Parameters=2.36M2024.05 | 96.22 | 90.3 | 90.77 | 94.69 | 89.91 | 85.66 | 91.26 | |
| FFTTuning=Full Fine-Tuning, Tunable Parameters=355M2024.05 | 95.99 | 90.4 | 90.81 | 94.6 | 90.39 | 85.92 | 91.35 | |
| LoPATuning=Low-Rank Prompt Adaptation, Tunable Parameters=1.60M2024.05 | 95.99 | 89.22 | 91.09 | 93.74 | 89.72 | 83.39 | 90.53 | |
| S-IDPGTuning=Prompt Tuning, Tunable Parameters=2.89M2024.05 | 95.3 | 84.5 | 78.6 | 90.48 | 84.88 | 77.26 | 85.17 | |
| prefix-tuningTuning=PEFT, Tunable Parameters=25.75M2024.05 | 93.8 | 89.51 | 90.86 | 94.98 | 86.87 | 82.67 | 89.78 | |
| p-tuning-v2Tuning=PEFT, Tunable Parameters=0.49M2024.05 | 89.91 | 88.13 | 70.18 | 85.21 | 84.63 | 53.43 | 78.58 | |
| DePTTuning=Prompt Tuning, Tunable Parameters=10.2K2024.05 | 89.68 | 71.51 | 72.97 | 57.09 | 45.81 | 53.79 | 65.14 | |
| Prompt Tuning (PT)Tuning=Prompt Tuning, Tunable Parameters=10.2K2024.05 | 84.4 | 54.67 | 72.38 | 58.74 | 48.2 | 53.07 | 61.91 | |
| Zero-shotTuning=None, Tunable Parameters=02024.05 | 59.97 | 39.6 | 73.52 | 50.16 | 42.34 | 53.43 | 53.17 |