Inspect Evals — запуск и разработка оценок LLM

★ 7.6 · vibe-coding

inspect_evals is a Claude Code skill that helps work with the Inspect Evals repository — a community-contributed collection of LLM evaluations built on the Inspect AI framework, developed jointly by UK AISI, Arcadia Impact, and the Vector Institute. The repository provides ready-to-run evals for benchmarking language models; installation is done via `pip install inspect-evals` or `uv sync`, and running an eval is as simple as `uv run inspect eval inspect_evals/arc_easy --model openai/gpt-5-nano`. Multiple tasks can be run simultaneously with `inspect eval-set`, and results are reviewed via `inspect view` or the Inspect AI VS Code extension. Supported model providers include OpenAI and Anthropic. The skill is aimed at AI safety researchers and ML engineers looking to evaluate models or contribute new evals to the register via a GitHub issue submission flow.