Skip to navigation

nemo_gym.verification

View as Markdown

Helpers that resources servers share for verifying submissions.

judge calls an LLM judge and routes a failed judge call to the failures sidecar (call_judge, judge_failsafe); verifier_fixture runs in-process contract tests of a server’s verifier; and sandbox_verification grades a submission in a fresh, single-use sandbox. sandbox_verification uses only the provider-neutral API; core sandbox APIs and providers live in nemo_gym.sandbox.

Submodules

Package Contents

Data

_PUBLIC_API

API

nemo_gym.verification._PUBLIC_API = {'JUDGE_FAILURES': 'judge', 'JudgeError': 'judge', 'JudgeGrading': 'judge', 'RUN...