Gemma 2
Google DeepMind — Gemma
rich detail 22 method fields published
Evaluation
Baseline assurance evaluations (safety policy violation rate)
violation rate for safety policies (child sexual abuse and exploitation, PII, hate speech/harassment, dangerous/malicious content, sexually explicit content, medical advice against consensus) using a large number of synthetic adversarial user queries with human raters labelling answers as policy-violating or not
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.
Gemma 3
Google DeepMind — Gemma
rich detail 21 method fields published