Paraphrase Detection on QQP (Accuracy and F1)
79.2AccuracySe²
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Se²2024.02 | 79.2 | 75.6 | |
| UPRISE2024.02 | 77.9 | 74.3 | |
| Best-of-102024.02 | 63.2 | 53.9 | |
| AES2024.02 | 63 | — | |
| SBERT2024.02 | 58.1 | 55.9 | |
| Instructor2024.02 | 56.4 | 57.4 | |
| BM252024.02 | 55.3 | 55.7 | |
| Zero-shot2024.02 | 48.4 | 42.1 | |
| Random2024.02 | 45.5 | 47.2 | |
| LinUpperBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=02026.04 | 37.85 | — | |
| UniformBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=02026.04 | 36.88 | — | |
| DoReMiBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=4.92 × 10^192026.04 | 36.84 | — | |
| RegMixBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=3.072 × 10^182026.04 | 36.83 | — | |
| ADAPT-BM25Backbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=≪ 1.0 × 10^142026.04 | 36.82 | — | |
| ADAPTBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=≪ 1.1 × 10^152026.04 | 36.81 | — |