Hugging Face's logo

Datasets:
tuskanny
/
sparse-nq-splade-cocondenser

Tasks:
Text Retrieval
Modalities:
Tabular
Formats:
csv
Size:
10K - 100K
Tags:
sparse
learned-sparse-retrieval
splade
beir
nq
approximate-nearest-neighbor
Libraries:
Datasets
pandas
Polars
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
0
int64
0
3.45k
5
int64
0
2.68M
1
int64
1
10
20.98173
float64
9.65
36.6
0
0
2
19.84291
0
1
3
19.398996
0
2,596,226
4
15.151361
0
169,203
5
14.812791
0
1,679,392
6
14.710797
0
2,422,438
7
14.214635
0
1,457,453
8
14.134082
0
1,997,283
9
13.946137
0
2,422,435
10
13.197068
1
6
1
26.414223
1
1,446,378
2
24.360565
1
2,007,442
3
24.161303
1
1,977,253
4
24.108929
1
1,234,354
5
24.06468
1
827,276
6
23.986
1
1,581,773
7
23.502874
1
827,287
8
23.286352
1
7
9
23.215271
1
1,078,543
10
23.152847
2
10
1
27.374884
2
12
2
24.402105
2
15
3
23.861961
2
279,614
4
21.63427
2
474,244
5
21.270031
2
14
6
21.23699
2
16
7
20.563536
2
976,398
8
20.481462
2
13
9
20.354025
2
539,832
10
19.750217
3
18
1
33.043465
3
1,425,987
2
32.60538
3
1,906,130
3
32.240253
3
41
4
31.654629
3
1,906,127
5
31.353443
3
1,906,117
6
30.854298
3
1,425,996
7
29.743155
3
1,425,988
8
29.621147
3
1,906,116
9
29.406868
3
1,425,994
10
29.204498
4
42
1
29.382807
4
1,808,792
2
27.277
4
1,808,790
3
26.98249
4
1,808,802
4
25.419085
4
1,808,795
5
20.94577
4
1,808,784
6
20.883972
4
1,808,786
7
20.313278
4
1,808,780
8
20.294651
4
1,808,788
9
20.149239
4
1,808,776
10
20.094866
5
47
1
23.489529
5
731,471
2
23.338322
5
50
3
22.517204
5
46
4
22.191795
5
58
5
21.935015
5
51,158
6
21.646608
5
1,724,829
7
21.459536
5
102,268
8
21.425385
5
686,490
9
20.94739
5
686,455
10
20.768623
6
2,354,373
1
18.488714
6
2,354,442
2
18.352482
6
1,238,292
3
18.145954
6
1,988,078
4
17.781017
6
2,354,439
5
17.054296
6
2,354,458
6
16.640835
6
2,354,462
7
16.277782
6
63
8
16.214993
6
462,491
9
16.214249
6
927,805
10
15.931425
7
821,318
1
21.189732
7
1,302,931
2
20.907024
7
76
3
19.998951
7
2,456,084
4
19.955648
7
78
5
19.836779
7
2,456,085
6
19.194773
7
67
7
18.969105
7
2,574,115
8
18.606844
7
66
9
18.57175
7
2,456,092
10
18.344648
8
244,986
1
17.295376
8
1,521,734
2
15.565739
8
244,959
3
15.473276
8
237,281
4
15.428103
8
244,995
5
15.278525
8
244,976
6
15.01251
8
244,960
7
14.682497
8
1,749,309
8
14.468931
8
86
9
14.316274
8
244,987
10
14.276554
9
91
1
23.850203
9
586,083
2
18.767172
9
928,780
3
18.407976
9
1,779,676
4
17.590513
9
1,236,506
5
17.555637
9
1,358,772
6
17.442705
9
586,089
7
17.410002
9
94
8
16.997427
9
104
9
16.814161
9
403,297
10
16.808376
10
558,461
1
30.163507
End of preview. Expand in Data Studio

Natural Questions (BEIR), SPLADE CoCondenser

Learned sparse embeddings of the BEIR version of Natural Questions (Wikipedia passages, Google queries, test split), encoded with SPLADE CoCondenser, ready for approximate nearest neighbor search by inner product.

Source

Statistics

Documents 2,680,893
Queries 3,452
Avg nonzeros per document 153.6
Avg nonzeros per query 50.7

Files

File Size Content
documents.bin 3.3 GB Document vectors
queries.bin 1.4 MB Query vectors
groundtruth.tsv 809 kB Exact top-10 per query
doc_ids.npy 2.8 GB BEIR document id of each document (string)
queries_ids.npy 111 kB BEIR query id of each query (string)

Format

Seismic binary format, little-endian, used by Seismic, kANNolo and Vectorium: u32 n_vectors, then for each vector u32 nnz, nnz x u32 component ids (ascending), nnz x f32 values.

groundtruth.tsv has four tab-separated columns: query position, document position, rank (1-based), score. Positions are 0-based rows of queries.bin and documents.bin. It holds the exact top-10 of every query by inner product, computed with an exhaustive scan. Row i of doc_ids.npy / queries_ids.npy is the original id of row i of documents.bin / queries.bin.

Component ids are integer term ids; no term mapping is included.
Downloads last month
33
Number of rows:
34,519
Total file size:
6.09 GB

Collection including tuskanny/sparse-nq-splade-cocondenser