Skip to navigation

nemo_gym.eval.comparison.suite.failures

View as Markdown

Read saved run failure accounting without scanning rollout records or attempt history.

Module Contents

Functions

NameDescription
read_saved_failuresRead this rollout file’s existing coverage snapshot; absent accounting remains unknown.
saved_failure_warningSummarize positive saved counts or unavailable accounting in one advisory line.

Data

_COUNT_KEYS

API

nemo_gym.eval.comparison.suite.failures.read_saved_failures(
path: pathlib.Path

Read this rollout file’s existing coverage snapshot; absent accounting remains unknown.

The saved counts cover the entire rollout file, including every agent and benchmark. This reader neither refreshes the snapshot nor interprets historical failure rows.

nemo_gym.eval.comparison.suite.failures.saved_failure_warning(
) -> str | None

Summarize positive saved counts or unavailable accounting in one advisory line.

nemo_gym.eval.comparison.suite.failures._COUNT_KEYS = {'total': 'failed', 'final': 'failed_final', 'not_final': 'failed_not_final', 'b...