Common Sense Reasoning on HellaSwag (dev)
95.4AccuracyDeBERTa-ASA
Evaluation results
| Method | Links | |
|---|---|---|
| DeBERTa-ASABackbone=DeBERTa-large, tau=0.3, runs=52022.06 | 95.4 | |
| DeBERTa (large)Backbone=DeBERTa-large, tau=0.3, runs=52022.06 | 94.3 | |
| Megatron-NLGModel Size=530B, Evaluation Protocol=Few-shot2021.12 | 82.4 | |
| GPT-3Model Size=175B, Evaluation Protocol=Few-shot, Shots=202021.12 | 79.3 | |
| GopherModel Size=280B, Evaluation Protocol=Few-shot, Shots=02021.12 | 79.2 | |
| GPT-3Model Size=175B, Evaluation Protocol=Zero-shot2021.12 | 78.9 | |
| GPT-3Model Size=175B, Evaluation Protocol=One-shot2021.12 | 78.1 | |
| GLaMModel Size=64B/64E, Evaluation Protocol=Few-shot, Shots=82021.12 | 77.2 | |
| GLaMModel Size=64B/64E, Evaluation Protocol=One-shot2021.12 | 76.8 | |
| GLaMModel Size=64B/64E, Evaluation Protocol=Zero-shot2021.12 | 76.6 | |
| Mamba2Discovery Source=Human Discovered2026.03 | 58.58 | |
| FusionGated-FIRNetDiscovery Source=AI Discovered2026.03 | 58.47 | |
| Gated DeltaNetDiscovery Source=Human Discovered2026.03 | 57.55 | |
| AdaMulti-PathGateNetDiscovery Source=AI Discovered2026.03 | 57.17 | |
| Content-SharpRouterDiscovery Source=AI Discovered2026.03 | 57 | |
| PathGate-FusionNetDiscovery Source=AI Discovered2026.03 | 56.99 | |
| Hier-GateNetDiscovery Source=AI Discovered2026.03 | 56.85 | |
| DeltaNetDiscovery Source=Human Discovered2026.03 | 56.29 | |
| BERT-ASABackbone=BERT-base, tau=0.3, runs=52022.06 | 40.8 | |
| BERTBackbone=BERT-base, tau=0.3, runs=52022.06 | 39.7 |