Safety Evaluation on Safety Benchmarks Aggregate (test)
73.6Generation Quality (Std Prefix)Self-Improving Pretraining
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Self-Improving PretrainingBackbone=Llama 1.4B, Training Dataset=RedPajama, Pretraining Objective=Safety2026.01 | 73.6 | 77.7 | 49.1 | 91.1 | |
| Llama Pretrain BaselineBackbone=Llama 1.4B, Training Dataset=RedPajama, Pretraining Objective=Next Token Prediction2026.01 | 54.5 | 52.6 | 47.9 | 75.5 | |
| Llama BaseBackbone=Llama 1.4B, Pretraining Objective=Base2026.01 | 50 | 50 | 47.6 | 76.9 | |
| Llama Pretrain BaselineBackbone=Llama 1.4B, Training Dataset=SlimPajama, Pretraining Objective=Next Token Prediction2026.01 | 49 | 44.9 | 46.8 | 77 |