Skip to content

Latest commit

 

History

History
28 lines (18 loc) · 1.12 KB

File metadata and controls

28 lines (18 loc) · 1.12 KB

medarc-verifiers

Utilities and CLI for running medical LLM benchmarks with verifiers. Provides TOML bench execution, result processing, and shared building blocks for authoring environments.

Install

pip install medarc-verifiers

Environments are installed separately via prime env install <owner/env> (from the Prime Intellect Hub) or vf-install <env> (from a local directory).

medarc-eval

medarc-eval covers the full evaluation pipeline:

Command Description
medarc-eval <ENV> Run a single benchmark; env-specific flags inferred from load_environment()
medarc-eval bench Run upstream TOML eval configs with deterministic MedARC paths
medarc-eval process Convert eval outputs to analysis-ready parquet
medarc-eval winrate Compute HELM-style win rates across models

See medarc-eval.md for full documentation.

Developer workflow

See developer-guide.md for local setup, environment authoring, and development workflow notes.