nemo_gym.utils.path_utils

Module Contents

Functions

Name	Description
`aggregate_metrics_path_for`	`results/rollouts.jsonl` -> `results/rollouts_aggregate_metrics.json`.
`aggregation_sources_path_for`	`results/rollouts.jsonl` -> `results/rollouts_sources.json`: the shards `gym eval aggregate` merged into it.
`attempt_extension_path_for`	Where `gym eval resubmit --recover --extend-attempts N` records the attempts it added to every rollout’s cap.
`batch_status_path_for`	`results/math.jsonl` -> `results/math_batch_status.json`.
`export_done_path_for`	`results/rollouts.jsonl`, `mlflow` -> `results/rollouts_mlflow_export_done`: set once the run was exported.
`exporter_run_id_path_for`	`results/rollouts.jsonl`, `mlflow` -> `results/rollouts_mlflow_run_id`: the run a live exporter logs to.
`failures_path_for`	-
`materialized_path_for`	`results/rollouts.jsonl` -> `results/rollouts_materialized_inputs.jsonl`.
`rerun_masked_journal_path_for`	Where `gym eval resubmit --recover --rerun-masked-without-artifact` journals the operator’s approval to run again
`resolved_config_path_for`	`results/rollouts.jsonl` -> `results/rollouts_resolved_config.yaml`.
`resubmit_offset_path_for`	Where `gym eval resubmit` records the failures sidecar’s line count; failures before it no longer count.
`run_summary_path_for`	`results/rollouts.jsonl` -> `results/rollouts_run_summary.json`.

Data

RERUN_MASKED_JOURNAL_NAME

API

nemo_gym.utils.path_utils.aggregate_metrics_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

results/rollouts.jsonl -> results/rollouts_aggregate_metrics.json.

Mirrors how rollout collection and reverification name the file they write, so consumers (e.g. gym eval compare) derive the same path the writers produced.

nemo_gym.utils.path_utils.aggregation_sources_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

results/rollouts.jsonl -> results/rollouts_sources.json: the shards gym eval aggregate merged into it.

nemo_gym.utils.path_utils.attempt_extension_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

Where gym eval resubmit --recover --extend-attempts N records the attempts it added to every rollout’s cap.

nemo_gym.utils.path_utils.batch_status_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

results/math.jsonl -> results/math_batch_status.json.

nemo_gym.utils.path_utils.export_done_path_for(
    output_fpath: pathlib.Path,
    exporter: str
) -> pathlib.Path

results/rollouts.jsonl, mlflow -> results/rollouts_mlflow_export_done: set once the run was exported.

nemo_gym.utils.path_utils.exporter_run_id_path_for(
    output_fpath: pathlib.Path,
    exporter: str
) -> pathlib.Path

results/rollouts.jsonl, mlflow -> results/rollouts_mlflow_run_id: the run a live exporter logs to.

nemo_gym.utils.path_utils.failures_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

nemo_gym.utils.path_utils.materialized_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

results/rollouts.jsonl -> results/rollouts_materialized_inputs.jsonl.

nemo_gym.utils.path_utils.rerun_masked_journal_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

Where gym eval resubmit --recover --rerun-masked-without-artifact journals the operator’s approval to run again masked results that kept a saved answer but have neither a gradeable artifact nor a grade: beside the rollouts file and named after it, like the attempt extension, so two rollouts files in one directory never share approvals.

results/rollouts.jsonl -> results/rollouts_rerun_masked_journal.jsonl.

nemo_gym.utils.path_utils.resolved_config_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

results/rollouts.jsonl -> results/rollouts_resolved_config.yaml.

nemo_gym.utils.path_utils.resubmit_offset_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

Where gym eval resubmit records the failures sidecar’s line count; failures before it no longer count.

nemo_gym.utils.path_utils.run_summary_path_for(
    output_fpath: pathlib.Path
) -> pathlib.Path

results/rollouts.jsonl -> results/rollouts_run_summary.json.

The per-run summary a Harbor run writes beside its rollouts (nemo_gym.data.tasks.harbor.run_summary), named like the other sidecars so a renamed --output keeps them together.

nemo_gym.utils.path_utils.RERUN_MASKED_JOURNAL_NAME = rerun_masked_journal_path_for(Path('rollouts.jsonl')).name