Skip to content

Evaluation

Safety overall performance (violation and false refusal rates)

Final violation rate (VR) and false refusal rate (FRR) of Llama 3 405B vs similar models (two end-to-end API competitor systems and one internally hosted open-source model); the Llama 3 model card additionally notes internal false-refusal benchmarks and mitigations to make Llama 3 significantly less likely to falsely refuse than Llama 2

Harmful content & safety Meta

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.

Llama 3
Meta

partial detail 12 method fields published

See how this test was run