harness: Add Vidore V3 benchmark and BEIR metrics support#1378
Open
jioffe502 wants to merge 8 commits intoNVIDIA:mainfrom
Open
harness: Add Vidore V3 benchmark and BEIR metrics support#1378jioffe502 wants to merge 8 commits intoNVIDIA:mainfrom
jioffe502 wants to merge 8 commits intoNVIDIA:mainfrom
Conversation
jioffe502
requested review from
ChrisJar,
charlesbluca and
drobison00
and removed request for
drobison00
- Add 8 Vidore V3 dataset configurations (finance_en, industrial, computer_science, pharmaceuticals, hr, energy, physics, finance_fr) - Add vidore_load_ground_truth() using HuggingFace datasets API - Add vidore_recall() evaluator with PDF-only matching - Add extract_page_as_image, extract_method, image_elements_modality config options to support Vidore's OCR-based page image retrieval - Add datasets>=2.0.0 dependency for HuggingFace qrels loading Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> Signed-off-by: Jacob Ioffe <jioffe@nvidia.com>
- Add dataset_groups section to test_configs.yaml with vidore, vidore_english, vidore_quick groups - Add expand_dataset_names() in config.py to handle group expansion - Add --list-datasets CLI option to show available datasets and groups - Update README.md with dataset groups documentation Usage: uv run nv-ingest-harness-run --list-datasets uv run nv-ingest-harness-run --case=e2e_recall --dataset=vidore uv run nv-ingest-harness-run --case=e2e_recall --dataset=vidore_quick Note: test_configs.yaml includes temp test settings (vdb_backend: milvus, reranker_mode: none, modified vidore_quick) - revert after testing. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add optional BEIR evaluation (NDCG, MAP, Precision) to recall tests - Configurable via enable_beir in test_configs.yaml or ENABLE_BEIR env var - Add beir>=2.0.0 dependency to harness - Add nvidia/llama-nemotron-embed-vl-1b-v2 to known embedding models Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add embed model fallback detection (dim=1024 warning) to e2e.py and recall.py - Add Milvus collection vector dimension verification after ingestion - Enable BEIR metrics by default for all Vidore V3 datasets Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: Jacob Ioffe <jioffe@nvidia.com>
jioffe502
force-pushed
the
vidore-v3-benchmark
branch
from
f4b0c94 to
e8af1a5
Compare
Aligns harness evaluation with the Vidore V3 notebook ground truth: - vidore_load_ground_truth now builds full qrels with all relevant docs and graded relevance scores (1=partial, 2=high) instead of collapsing to single doc with binary relevance - Dedup retrieved PDFs in recall scoring and BEIR metrics to avoid multiple chunks from the same PDF inflating top-k positions - Add language_filter config for isolating English-only query evaluation - Add warm-up sleep and TeeFile close guard Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: Jacob Ioffe <jioffe@nvidia.com>
jioffe502
changed the title
ChrisJar
approved these changes
Feb 25, 2026
Collaborator
ChrisJar
left a comment
ChrisJar
left a comment
There was a problem hiding this comment.
Thanks for working on this!
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Adds Vidore V3 benchmark support and BEIR evaluation metrics to the test harness.
Changes
--dataset=vidore)results in: https://docs.google.com/spreadsheets/d/137poeB7CmDE7AmaiLalOM1qZZvDlCDRkBfMD6IjBBAs/edit?gid=0#gid=0
Dependent on #1305
Checklist