Dataset Viewer
Auto-converted to Parquet Duplicate
query-id
stringlengths
24
26
corpus-id
stringlengths
26
28
score
int64
1
1
meaning-q-2c42686e876a5257
meaning-doc-1cb6dd6fb4a2130d
1
meaning-q-9f625920dfd4e5c0
meaning-doc-94fe659e68f032ad
1
meaning-q-d0fc89caf7c2fb65
meaning-doc-ece9c42a86c77ac8
1
meaning-q-669c18c6d9ceb242
meaning-doc-6240a98eb98afcbe
1
meaning-q-4c9f04173106181f
meaning-doc-4cd77d54ba799cd3
1
meaning-q-f9917043e0bf1f61
meaning-doc-212e6f3acaa32c60
1
meaning-q-51b0ec70c5a3374b
meaning-doc-319935123d25728c
1
meaning-q-0787509db645d64b
meaning-doc-d7719615ba1fdb91
1
meaning-q-dd78a790212a5cf3
meaning-doc-ede68fe431bb5062
1
meaning-q-150965e13488196f
meaning-doc-8e7926e122b37dd9
1
meaning-q-aced6c00a7deac88
meaning-doc-6a1e0c70f25de37c
1
meaning-q-14bb49b73af1f903
meaning-doc-b6bdcb354f45c4c7
1
meaning-q-626cb063daec6b01
meaning-doc-7046641939de102c
1
meaning-q-9c19bc5c201b5f57
meaning-doc-109427e1acf11297
1
meaning-q-2a5125c6af083e28
meaning-doc-c818ccfeae0d3dec
1
meaning-q-29c92422bd528b09
meaning-doc-667ca081762adeaa
1
meaning-q-96e4c0fb458dc1a1
meaning-doc-5f3d39846c5f2554
1
meaning-q-6c1e5795031167ac
meaning-doc-18621fe31cfb50cb
1
meaning-q-645ff74d36e6b142
meaning-doc-1715d58110f3632c
1
meaning-q-a8c079362004ae8d
meaning-doc-3fbd4362b259c3e2
1
meaning-q-3ebefa59e25c8baa
meaning-doc-f3aa76cf83c8a229
1
meaning-q-b10286905dc6d80c
meaning-doc-dd82652b5ae49aae
1
meaning-q-cd35f49e84f33010
meaning-doc-ad3dc8e93d423d85
1
meaning-q-8d6a9e0f369c41aa
meaning-doc-1908189a7a729225
1
meaning-q-4b0fa21c23ace07e
meaning-doc-3a701a1f8894b13f
1
meaning-q-90338ad7316ef7e5
meaning-doc-f312b31486a8fbb2
1
meaning-q-591f6de3bc942cb0
meaning-doc-a381bc2027d02230
1
meaning-q-ceabc9b184e27850
meaning-doc-a854de5f0f212325
1
meaning-q-e64293bf0799655b
meaning-doc-344feb22ce5ba5e5
1
meaning-q-482a7b1ebdd41b17
meaning-doc-67c6598ba70d6843
1
meaning-q-3c4b715327305bd2
meaning-doc-9841beb080766366
1
meaning-q-e950d6a09442637f
meaning-doc-2bbe9e55eb6d5082
1
meaning-q-4661e56f1bbc3a1c
meaning-doc-a8cbee365871fa62
1
meaning-q-5f981fa102f182f8
meaning-doc-1276954434968fd0
1
meaning-q-7da673696249c1ad
meaning-doc-3786ec3425f4ce14
1
meaning-q-7be9aee10c466bc4
meaning-doc-dded13f1cba8f993
1
meaning-q-e7108720fc1dc882
meaning-doc-6a18a41a125f97a8
1
meaning-q-9e6f7c79411ae25b
meaning-doc-8ad6e24041ae7ec7
1
meaning-q-4b603ae1a3f171ef
meaning-doc-d9b9c573916c1040
1
meaning-q-81d5b683ec224f22
meaning-doc-be51db78748d7c4f
1
meaning-q-5580fa6d28d0689b
meaning-doc-28dbf320f83e3f1f
1
meaning-q-6f238a99e79a0680
meaning-doc-f55b7448e01a04c6
1
meaning-q-7811f27b701cd130
meaning-doc-ccaa96eb0f33629a
1
meaning-q-ae7bda0ee7702a65
meaning-doc-43aa9bf57c67f8f5
1
meaning-q-01f23a04064b9403
meaning-doc-4a59b06faaf7b2ed
1
meaning-q-55efae4da23d0634
meaning-doc-fd61fe9b3e9a0472
1
meaning-q-e21d8bbe4f8b979e
meaning-doc-6247631d343a08a9
1
meaning-q-ccb3b75f7a6fdbc4
meaning-doc-bb45fbbd907076a0
1
meaning-q-fd3439e5373bfa94
meaning-doc-1d4a1922259d714f
1
meaning-q-b04b99eecc5580d9
meaning-doc-05b46fb36b2a775a
1
meaning-q-47906031aa1d62f1
meaning-doc-591e6712d5733f46
1
meaning-q-e016e960d431deb3
meaning-doc-04b728af427ac43f
1
meaning-q-ae58d1595bfb05d6
meaning-doc-5881293d73885516
1
meaning-q-8a4b646f6acfe4d1
meaning-doc-0263584d9b993949
1
meaning-q-44493b52ae475fcb
meaning-doc-9761f82d20a333a1
1
meaning-q-91740265a5b106e4
meaning-doc-3ec21618656bb3e0
1
meaning-q-396b9667987543cf
meaning-doc-c4469aa156f2691a
1
meaning-q-28cdd2b1ee0ac88c
meaning-doc-3872e18aa0675e32
1
meaning-q-d2bd62209222f0c5
meaning-doc-0021aa6d4b4f49a2
1
meaning-q-e19505456ba968c6
meaning-doc-c829fcf6a2019ad5
1
meaning-q-2af2d671151b5812
meaning-doc-46a9e139f4df2fdf
1
meaning-q-5c4b0a3fe10da5c5
meaning-doc-719d438f3d58ca3b
1
meaning-q-47200b6c3c26f49b
meaning-doc-8a285580a2fb13f0
1
meaning-q-c32e1a097facd50c
meaning-doc-7c7e2adfe7854628
1
meaning-q-cda180348e3abdc6
meaning-doc-81dd2c1029580dce
1
meaning-q-987cff8c7715d5ce
meaning-doc-7a8c2ab0a8836a88
1
meaning-q-f41ff99abcb89b89
meaning-doc-d46ca36833dcc4c7
1
meaning-q-3bd3752a360ba0ab
meaning-doc-e73a97040e513fa7
1
meaning-q-10972b94d674c2bc
meaning-doc-a29b5799224c0529
1
meaning-q-76ba86a4dd895231
meaning-doc-c9c541db38a8c172
1
meaning-q-f56bcc3204005c96
meaning-doc-15aeccddb27426b9
1
meaning-q-a745463216df5e17
meaning-doc-bccd04b93eb69e14
1
meaning-q-f086e2e78eb0bc60
meaning-doc-b9d158b173db19a2
1
meaning-q-5ebb795e549402d6
meaning-doc-f34a5eb6186d8b2c
1
meaning-q-5da69b380857d131
meaning-doc-c24ca05ef2481050
1
meaning-q-c4b3825761b444b6
meaning-doc-260db72e775d23ac
1
meaning-q-dc90a786320ba575
meaning-doc-fdb116e18564641f
1
meaning-q-ef114b34bb24d190
meaning-doc-9a4ab7678a96d4e2
1
meaning-q-f5774541b3644860
meaning-doc-085d57e38d59db91
1
meaning-q-4958e38b9c2cd08f
meaning-doc-1057ddfc826c297c
1
meaning-q-a0962e388d7b3a68
meaning-doc-1945eb83bf85962d
1
meaning-q-27978771f06afec5
meaning-doc-5423f00a00d53ae9
1
meaning-q-1e6c7cee97ebd013
meaning-doc-0d90e29687e940e6
1
meaning-q-f0bff7147c40e626
meaning-doc-28dac5dfd6693aec
1
meaning-q-af00683299f251ed
meaning-doc-69a21bb8acc7c941
1
meaning-q-875b4f5fdca4aea5
meaning-doc-622c6b8a7021f547
1
meaning-q-7a0ee50a790c342c
meaning-doc-5cc2cba9c577526b
1
meaning-q-9e11b7a9b9be4e10
meaning-doc-f09e6bb2cce02167
1
meaning-q-b48079bde7b1e337
meaning-doc-0ce13f7b59de99a6
1
meaning-q-3be6dee3f39fee10
meaning-doc-ab61c9911879e3e5
1
meaning-q-fd0192fac012074c
meaning-doc-eef6d1152cc4c91d
1
meaning-q-fb653419ffd5b450
meaning-doc-b7e5292109226982
1
meaning-q-0dcf1c0c837ed019
meaning-doc-a475c27a06f0dd72
1
meaning-q-dd1a0dc515b4cdb5
meaning-doc-f02ccb85427f1af9
1
meaning-q-5cb36fc589b939fb
meaning-doc-15ca8d7bdbd5ca26
1
meaning-q-d32d9a814db22d24
meaning-doc-28f3d3be0620f387
1
meaning-q-8236140131626908
meaning-doc-667505a83c3083f4
1
meaning-q-f13cb7a29d08c59d
meaning-doc-bfacfa9046f08b37
1
meaning-q-0914f92a77b5645a
meaning-doc-2a6bd70e5cad5cc1
1
meaning-q-a4af93f74f7f3ae7
meaning-doc-84b5e79e64f6bc15
1
End of preview. Expand in Data Studio

ParseEmbed

Hard, parse-sensitive retrieval evaluation for embedding models.

ParseEmbed is a compact benchmark for embedding models. It tests whether a model can retrieve the exact correct document when hard negatives share nearly all surface tokens with the answer.

Tasks

Task ID Split What it measures
mean mean Semantic scope, negation, numeric values, temporal conditions, and exception handling
text_formatting text_formatting Meaning carried by Markdown-like formatting such as code spans, headings, quotes, and strike-through
table table Row/column grounding in compact Markdown tables

Files

  • eval.yaml: Hugging Face Benchmark definition.
  • mean.jsonl, text_formatting.jsonl, table.jsonl: benchmark splits.
  • corpus.jsonl: retrieval corpus.
  • queries.jsonl: retrieval queries.
  • qrels.jsonl: relevance judgments.
  • evaluate_parseembed.py: reference evaluator for embedding models.
  • requirements.txt: evaluator dependencies.

Quick Start

Install the evaluator dependencies:

pip install -r requirements.txt

Run a local evaluation:

python evaluate_parseembed.py   --model sentence-transformers/all-MiniLM-L6-v2   --task all   --k 10   --output results.json

The evaluator reports ndcg_at_10, recall_at_10, and mrr_at_10 for each task and an average across tasks.

Evaluation Protocol

ParseEmbed is evaluated as dense retrieval:

  1. Embed every document in corpus.jsonl.
  2. Embed the queries from one benchmark split.
  3. Rank corpus documents by cosine similarity.
  4. Score the rank of each query's positive_doc_id.

The official leaderboard metric should be ndcg_at_10. recall_at_10 and mrr_at_10 are included as secondary diagnostics.

Data Format

Each benchmark split contains one JSON object per query with these fields: id, query, positive_doc_id, positive_text, hard_negative_doc_ids, hard_negative_texts, answer, style, and difficulty.

The corpus is stored separately in corpus.jsonl, and qrels.jsonl provides standard retrieval relevance judgments with query-id, corpus-id, and score.

Eval Results

Model repositories can report results with .eval_results/parseembed.yaml:

- dataset:
    id: Convence/ParseEmbed
    task_id: mean
    revision: <dataset_commit_hash>
  value: <ndcg_at_10_score>
  notes: "metric=ndcg_at_10"

Use task IDs mean, text_formatting, and table.

Benchmark Registration

This repository is prepared for Hugging Face's beta Benchmark system:

  • It contains a root eval.yaml.
  • evaluation_framework is set to parseembed.
  • The benchmark has three task leaderboards: mean, text_formatting, and table.
  • A working reference evaluator is included in evaluate_parseembed.py.

Because Hugging Face's Benchmark feature is allow-listed, parseembed must be added by the Hugging Face team before the official Benchmark tag appears.

Usage

from datasets import load_dataset

mean = load_dataset("Convence/ParseEmbed", "parse-embed", split="mean")
corpus = load_dataset("Convence/ParseEmbed", "corpus", split="test")
queries = load_dataset("Convence/ParseEmbed", "queries", split="test")
qrels = load_dataset("Convence/ParseEmbed", split="test")

Dataset Construction

The benchmark is generated deterministically from generate_parseembed.py with seed 91247. Each query has one positive document and three hard negatives. The negatives are constructed by changing a decisive detail while preserving most of the vocabulary.

No private or external source text is used. The benchmark is synthetic and is intended as a diagnostic stress test alongside natural retrieval benchmarks.

Reproducibility

ParseEmbed is fully deterministic. Regenerating from generate_parseembed.py with the same seed produces the same queries, corpus IDs, hard negatives, and relevance judgments.

Size

  • Queries: 720
  • Corpus documents: 2,880
  • Queries per task: 240

License

Apache 2.0

Downloads last month
21