Question Answering on TruthfulQA (MC1, MC2 Metrics)
27.7MC1Self-Improving Pretraining
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Self-Improving PretrainingTraining Data=SlimPajama, Pretraining for=Factuality2026.01 | 27.7 | 42.5 | |
| Llama BaseStage=Base Model2026.01 | 22.4 | 35.9 | |
| Llama Pretrain BaselineTraining Data=SlimPajama, Pretraining for=Factuality2026.01 | 21.5 | 35.5 |