Question Answering on ARC Challenge (Acc. and Token)
96.7AccuracyLaTER (training)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LaTER (training)Backbone=Qwen3-14B2026.05 | 96.7 | 434 | |
| LaTER (training-free)Backbone=Qwen3-14B2026.05 | 96.5 | 533 | |
| CoT BaselineBackbone=Qwen3-14B2026.05 | 96.3 | 631 | |
| CoT-SFTBackbone=Qwen3-14B2026.05 | 95.5 | 578 |