Question Answering on SQuAD short (EM)
64.2Exact Match (EM)GPT-2-124m
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-2-124mcontext_type=Full context models (upper bound)2026.03 | 64.2 | |
| Mamba-130m-hfcontext_type=Recurrent models2026.03 | 63.3 | |
| GradMem (GPT-2, increased K)K=increased, Number of memory tokens=322026.03 | 54.9 | |
| Pythia-160mcontext_type=Full context models (upper bound)2026.03 | 48.9 | |
| RMT (GPT-2)context_type=Recurrent models2026.03 | 42.6 | |
| ARMT (GPT-2)context_type=Recurrent models2026.03 | 39 | |
| GradMem (GPT-2, K = 1)K=1, Number of memory tokens=322026.03 | 38.1 |