Natural Language Understanding on QNLI (Exact Match)
82.5Exact MatchAdapShot
Evaluation Results
| Method | Links | |
|---|---|---|
| AdapShotBackbone=Qwen2.5-7B2026.05 | 82.5 | |
| DBSABackbone=Qwen2.5-7B2026.05 | 80.7 | |
| Many-shot (256)Backbone=Qwen2.5-7B2026.05 | 78 | |
| Many-shot (512)Backbone=Qwen2.5-7B2026.05 | 76.8 | |
| Many-shot (1024)Backbone=Qwen2.5-7B2026.05 | 66.7 | |
| DBSABackbone=LLaMA-3.2 (3B)2026.05 | 61.9 | |
| Many-shot (512)Backbone=LLaMA-3.2 (3B)2026.05 | 57.9 | |
| Many-shot (256)Backbone=LLaMA-3.2 (3B)2026.05 | 55.7 | |
| Few-shot (8)Backbone=Qwen2.5-7B2026.05 | 54.1 | |
| Many-shot (1024)Backbone=LLaMA-3.2 (3B)2026.05 | 52.6 | |
| Few-shot (8)Backbone=LLaMA-3.2 (3B)2026.05 | 51.7 | |
| AdapShotBackbone=LLaMA-3.2 (3B)2026.05 | 51.2 | |
| Zero-shotBackbone=LLaMA-3.2 (3B)2026.05 | 50.8 | |
| Zero-shotBackbone=Qwen2.5-7B2026.05 | 20.3 |