Skip to content

Harness/retriever nightly refresh#1522

Merged
jdye64 merged 6 commits intoNVIDIA:mainfrom
jioffe502:harness/retriever-nightly-refresh
Mar 10, 2026
Merged

Harness/retriever nightly refresh#1522
jdye64 merged 6 commits intoNVIDIA:mainfrom
jioffe502:harness/retriever-nightly-refresh

Conversation

@jioffe502
Copy link
Collaborator

Description

Add a production-friendly nightly summary flow for the retriever harness, including Slack posting, replay support, and compact reporting for the metrics we actually monitor.

  • Add nightly Slack posting and replay support, with compact session_summary.json rollups for pages, ingest_secs, pages_per_sec_ingest, recall_5, and artifact_dir
  • Fix nested dataset discovery so directory layouts like earnings_consulting work correctly in nightly runs, and enable FinanceBench nightly recall with the repository fixture JSON
  • Consolidate recursive input discovery into a shared helper and clean up hard-coded user-specific dataset path examples in docs

Test plan

  • PYTHONPATH="/localhome/local-jioffe/nv-ingest-nightly-pr/nemo_retriever/src" "/localhome/local-jioffe/nv-ingest/.retriever/bin/python" -m pytest -q nemo_retriever/tests/test_batch_pipeline.py nemo_retriever/tests/test_harness_config.py nemo_retriever/tests/test_harness_run.py nemo_retriever/tests/test_harness_nightly.py
  • PYTHONPATH="/localhome/local-jioffe/nv-ingest-nightly-pr/nemo_retriever/src" "/localhome/local-jioffe/nv-ingest/.retriever/bin/python" -m nemo_retriever.harness.__main__ nightly --runs-config "/localhome/local-jioffe/nv-ingest-nightly-pr/nemo_retriever/harness/nightly_config.yaml" --dry-run

Checklist

  • I am familiar with the Contributing Guidelines.
  • New or existing tests cover these changes.
  • The documentation is up to date with these changes.
  • If adjusting docker-compose.yaml environment variables have you ensured those are mimicked in the Helm values.yaml file.

Add a production-friendly nightly summary flow so shared benchmark boxes can post compact retriever results to Slack and replay prior sessions without rerunning long jobs.
Consolidate recursive dataset input resolution into one shared helper so nightly and batch entrypoints stay aligned, and remove user-specific dataset path examples from docs.
@jioffe502 jioffe502 requested a review from a team as a code owner March 9, 2026 21:27
@jioffe502 jioffe502 requested a review from edknv March 9, 2026 21:27
Add reusable shell entrypoints for one-shot nightly runs and a tmux-friendly loop mode so benchmark boxes can reuse the same invocation path for scheduled and manual execution.
Avoid importing requests during harness module import so pytest collection works in minimal environments that never execute Slack posting.
Fold formatter and flake8-driven line wrapping into the nightly harness changes so the refreshed PR branch stays clean under the repo hooks.
@jdye64 jdye64 merged commit 739a193 into NVIDIA:main Mar 10, 2026
9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants