notesum.ai
Published at December 9A Comparative Study of Learning Paradigms in Large Language Models via Intrinsic Dimension
cs.CL
Released Date: December 9, 2024
Authors: Saahith Janapati1, Yangfeng Ji1
Aff.: 1University of Virginia

| Dataset | ICL-0 | ICL-1 | ICL-2 | ICL-5 | ICL-10 | Finetune 1K |
|---|---|---|---|---|---|---|
| SST-2 | 0.685 | 0.633 | 0.731 | 0.807 | 0.832 | 0.944 |
| CoLA | 0.720 | 0.723 | 0.735 | 0.746 | 0.742 | 0.750 |
| QNLI | 0.517 | 0.513 | 0.555 | 0.590 | 0.585 | 0.761 |
| QQP | 0.417 | 0.462 | 0.485 | 0.508 | 0.519 | 0.707 |
| MNLI | 0.374 | 0.367 | 0.387 | 0.414 | 0.431 | 0.676 |
| AGNews | 0.638 | 0.573 | 0.712 | 0.772 | 0.809 | 0.881 |
| CommonsenseQA | 0.199 | 0.375 | 0.417 | 0.470 | 0.492 | 0.500 |
| MMLU | 0.449 | 0.488 | 0.511 | 0.524 | 0.531 | 0.542 |