Llama 2 7B Chat

Evidence summary

Safety. Llama 2 7B Chat has an estimated Safety rank of #231; its 90% source-sensitivity interval is #140–#268. Its behavior-only rank is #218; company governance moves the combined estimate to #231. Published Safety evidence spans 16 eval lineages and 7 of 7 components. Its strongest relative result is MedSafetyBench (medical_safety_score, #1 of 30); its weakest is NESSiE Necessary Safety Benchmark (sh_score, #16 of 16).

Freedom. Llama 2 7B Chat has an estimated Freedom rank of #584; its 90% source-sensitivity interval is #301–#633. Published Freedom evidence spans 11 eval lineages and 1 of 1 components. Its strongest relative result is COMPL-AI LLM RuLES Multi-Turn Rule Following (score, #3 of 14); its weakest is Do-Not-Answer (human_harmlessness_rate, #6 of 6).