Question Answering on bAbI en
26.8Few-Shot AccuracyGPT-2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-2#Params=124M, Patch size=-, few-shot evaluation=true2026.04 | 26.8 | — | |
| MGPT-2#Params=124M, Patch size=-, few-shot evaluation=true2026.04 | 24.3 | — | |
| MIXAR stage2#Params=477M, Patch size=32×32, few-shot evaluation=true2026.04 | 22.5 | 61.4 | |
| PIXAR stage2#Params=113M, Patch size=8×8, few-shot evaluation=true2026.04 | 19.6 | 77 | |
| MIXAR stage2#Params=116M, Patch size=32×32, few-shot evaluation=true2026.04 | 16.8 | 72.7 | |
| MIXAR stage1#Params=477M, Patch size=32×32, few-shot evaluation=true2026.04 | 11.3 | 61.8 | |
| MIXAR stage1#Params=116M, Patch size=32×32, few-shot evaluation=true2026.04 | 7.4 | 52.6 |