nemo_gym.rollout_compare.compare

Classify every task across the two sides.

compare_rollouts detects the keys of each side, decides whether tasks are keyed by taskset, groups both sides, picks the join strategy, labels each task identical, flipped, flipped with an agent error, masked or missing, lists the anonymous sidecar failures, and returns a CompareResult with its notes.

Module Contents

Functions

Name	Description
`compare_rollouts`	-

API

nemo_gym.rollout_compare.compare_rollouts(
    old_rows: list[dict[str, typing.Any]],
    new_rows: list[dict[str, typing.Any]],
    key: typing.Optional[str] = None
) -> nemo_gym.rollout_compare.schema.CompareResult