Llama 3
Meta
partial detail 9 method fields published
Evaluation
Safety supervised finetuning evaluation (VR/FRR balance)
Safety SFT combining helpfulness and safety data (Llama 2 recipe) plus a borderline dataset to discern safe vs unsafe requests; results on balancing violation rate (VR) and false refusal rate (FRR) by model size
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.