Reading Comprehension on DROP Multiple Client (test) (Client Accuracies, Avg. Accuracy, PGR)
64.73Client 1 AccuracyPT
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| PTDescription=Ceiling performance, LLM fine-tuned directly on client data2026.05 | 64.73 | 64.08 | 60.98 | 63.82 | 66.02 | 63.93 | — | |
| GRAD-TRANSFORMERDescription=Learning to generate updates for LLMs2026.05 | 60.72 | 64.34 | 61.37 | 64.47 | 62.53 | 62.69 | 84.38 | |
| ConfDescription=Confidence-based weak-to-strong distillation2026.05 | 58.91 | 64.21 | 63.82 | 64.34 | 61.63 | 62.58 | 83 | |
| W2SDescription=Weak-to-strong distillation2026.05 | 57.24 | 61.24 | 61.63 | 63.57 | 61.63 | 61.06 | 63.85 | |
| PSDescription=TinyLM fine-tuned from the client side2026.05 | 56.98 | 55.81 | 53.36 | 58.91 | 54.91 | 55.99 | — | |
| VisSupDescription=Weak-to-strong distillation baseline2026.05 | 56.85 | 63.44 | 62.14 | 63.82 | 61.11 | 61.47 | 69.02 |