docs: fill CLI reference gaps for data prep and rollout collection#1498
docs: fill CLI reference gaps for data prep and rollout collection#1498lbliii wants to merge 7 commits into
Conversation
Add missing CLI reference coverage in cli-commands.mdx: - ng_materialize_prompts: full parameter table, example, and a note comparing it to ng_prepare_data (closes #1347). - resume_from_cache: params-table entries for ng_collect_rollouts and ng_e2e_collect_rollouts plus a "Resume interrupted runs" section covering materialized inputs, incremental flush, matching, and the stale-cache footgun (closes #1239). - Generation parameters: dedicated section documenting how temperature, top_p, and max_output_tokens are passed via ++responses_create_params.<field> rather than as standalone flags (closes #637). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: Lawrence Lane <llane@nvidia.com>
- "See" -> "Refer to" in the new cross-references (accessibility wording). - Title-case the two new subheadings (Generation Parameters, Resume Interrupted Runs); anchors are unchanged so links still resolve. - Drop the academic adverb "gracefully". Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: Lawrence Lane <llane@nvidia.com>
There was a problem hiding this comment.
Pull request overview
Updates the NeMo Gym Fern documentation CLI reference (fern/versions/latest/pages/reference/cli-commands.mdx) to close gaps around rollout collection resume behavior, generation parameter overrides, and the ng_materialize_prompts data-prep command.
Changes:
- Adds
resume_from_cacheto theng_collect_rolloutsandng_e2e_collect_rolloutsparameter tables and documents the resume workflow/artifacts. - Adds a Generation Parameters section clarifying how to override sampling params via
++responses_create_params.<field>. - Adds a full CLI reference entry for
ng_materialize_prompts, including parameters, example usage, and guidance vs.ng_prepare_data.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
|
|
||
| #### Generation Parameters | ||
|
|
||
| Sampling parameters such as `temperature`, `max_output_tokens`, and `top_p` are not standalone CLI flags — they are passed as overrides inside `responses_create_params` using Hydra's nested dot syntax. Each override is merged into every input row's existing `responses_create_params`: |
There was a problem hiding this comment.
Addressed in 91526fc — the Generation Parameters section now states that overrides use a shallow merge (top-level keys only), matching dict | overrides in _preprocess_rows_from_config.
| ++responses_create_params.max_output_tokens=4096 | ||
| ``` | ||
|
|
||
| The same syntax works for `ng_e2e_collect_rollouts`. Any field accepted by the Responses API create params can be set this way (for example, `++responses_create_params.reasoning.effort=low`). |
There was a problem hiding this comment.
Addressed in 91526fc — narrowed the nested-object guidance: top-level fields are straightforward, but nested Hydra overrides replace the entire nested dict (other fields under that key are not preserved).
|
🌿 Preview your docs: https://nvidia-preview-docs-cli-reference-cleanup.docs.buildwithfern.com/nemo/gym Here are the markdown pages you've updated: |
|
Closing in favor of #1675 — DCO was failing because commit |
Summary
Fills documented gaps in the CLI reference and installation pages. All CLI additions were verified against
nemo_gym/prompt.pyandnemo_gym/rollout_collection.py.ng_materialize_prompts— full parameter table, example, and a note clarifying when to use it vs.ng_prepare_data. Closes docs: add missing CLI commands to reference and clarify when to use each #1347.resume_from_cache— params-table entries forng_collect_rolloutsandng_e2e_collect_rollouts, plus a Resume Interrupted Runs section covering the*_materialized_inputs.jsonlartifact, per-rollout incremental flush,(task_index, rollout_index)matching, fallback behavior, and the stale-cache footgun. Closes docs: document resume_from_cache for rollout collection #1239.temperature,top_p, andmax_output_tokensare passed via++responses_create_params.<field>rather than as standalone flags. Documents shallow (top-level) merge behavior for nested overrides. Closes bug: could not find temperature CLI params due to docs segmented control #637.installation.mdx. Closes docs: swap install path to pip install nemo-gym with 0.3.0 PyPI release #1191.Review feedback addressed
responses_create_paramsoverrides use a shallow merge (dict | overridesin_preprocess_rows_from_config), so nested Hydra overrides replace the entire nested dict rather than deep-merging fields.Verification
npm run check(fern check) passes with 0 errors.