pip install lm-eval

Installs the lm-evaluation-harness package, which provides the `lm-eval` command for benchmarking language models on tasks like MMLU and hellaswag. The base install covers CPU evaluation; add extras for backends, for example `pip install lm-eval[hf]` or `pip install lm-eval[vllm]`, which pull torch or vLLM. PyPI's project name uses an underscore, but pip resolves lm-eval the same way.