lotte_pooled_colbertv2

0 int64 1 2.93k	Q0 stringclasses 1 value	9032 int64 4 2.43M	1 int64 1 1
1	Q0	39,506	1
1	Q0	39,507	1
2	Q0	4,893	1
2	Q0	4,895	1
2	Q0	4,912	1
3	Q0	31,507	1
3	Q0	90,314	1
4	Q0	29,672	1
4	Q0	29,678	1
4	Q0	29,691	1
5	Q0	1,248	1
5	Q0	1,252	1
5	Q0	1,257	1
5	Q0	1,265	1
5	Q0	100,587	1
6	Q0	5,024	1
6	Q0	26,234	1
7	Q0	32,543	1
7	Q0	32,552	1
8	Q0	36,810	1
9	Q0	10,980	1
9	Q0	10,994	1
9	Q0	11,018	1
10	Q0	16,889	1
10	Q0	95,961	1
11	Q0	10,199	1
11	Q0	10,201	1
12	Q0	47,006	1
12	Q0	47,009	1
12	Q0	47,010	1
13	Q0	1,512	1
14	Q0	42,618	1
14	Q0	42,630	1
15	Q0	3,375	1
15	Q0	40,181	1
16	Q0	21,453	1
16	Q0	82,134	1
17	Q0	12,442	1
18	Q0	55,389	1
18	Q0	55,394	1
19	Q0	13,445	1
20	Q0	1,495	1
20	Q0	6,133	1
20	Q0	23,580	1
21	Q0	24,466	1
21	Q0	24,467	1
22	Q0	9,493	1
22	Q0	9,494	1
23	Q0	59,903	1
23	Q0	59,904	1
23	Q0	59,908	1
24	Q0	81,132	1
25	Q0	78,026	1
25	Q0	78,028	1
25	Q0	78,036	1
26	Q0	6,092	1
26	Q0	60,542	1
27	Q0	21,819	1
27	Q0	21,902	1
27	Q0	21,921	1
27	Q0	22,050	1
28	Q0	75,437	1
29	Q0	77,361	1
29	Q0	77,365	1
29	Q0	77,392	1
30	Q0	26,611	1
30	Q0	26,613	1
31	Q0	42,463	1
31	Q0	42,479	1
31	Q0	42,480	1
31	Q0	42,485	1
31	Q0	42,486	1
31	Q0	42,582	1
32	Q0	14,268	1
32	Q0	14,270	1
32	Q0	14,283	1
32	Q0	17,713	1
32	Q0	22,662	1
32	Q0	71,090	1
33	Q0	58,928	1
34	Q0	7,697	1
34	Q0	7,698	1
35	Q0	3,820	1
35	Q0	11,832	1
35	Q0	21,275	1
35	Q0	21,548	1
35	Q0	106,764	1
36	Q0	20,789	1
37	Q0	44,214	1
38	Q0	4,565	1
38	Q0	4,579	1
39	Q0	49,661	1
39	Q0	49,663	1
39	Q0	56,916	1
40	Q0	68,195	1
41	Q0	7,706	1
42	Q0	62,838	1
42	Q0	62,839	1
42	Q0	62,840	1
42	Q0	62,846	1

End of preview. Expand in Data Studio

LoTTE pooled (dev), ColBERTv2

Token-level (late-interaction) ColBERTv2 embeddings of the LoTTE pooled dev collection and its search queries.

Source

Collection: LoTTE pooled, dev split (ir_datasets lotte/pooled/dev), 2,428,854 passages
Queries: dev search queries, 2,931 (lotte/pooled/dev/search, 8,573 qrels)
Document order: passage id order (row i is pid i)

Encoding

Model: ColBERTv2 (colbert-ir/colbertv2.0, BERT-base-uncased tokenizer)
Documents start with [CLS] and the ColBERT [D] marker (token id 2)
Document length cap: the longest document has 180 vectors, consistent with the ColBERTv2 default doc_maxlen=180
Queries: always 32 vectors ([MASK] expansion, no zero padding)
Vectors: 128-d, L2-normalized
Not recorded: exact checkpoint revision, encoding library, and whether punctuation was dropped

Statistics


Token vectors (N)	266,205,513
Avg vectors per document	109.6 (min 3, max 180)
Vectors per query	32

Files

File	dtype	Shape	Content
`documents.npy`	uint16 (`<u2`)	`[266205513, 128]`	Raw float16 bit patterns stored as uint16. Read with `.view(np.float16)`
`doclens.npy`	int32	`[2428854]`	Vectors per document; `sum == N`
`token_ids.npy`	int64	`[266205513]`	Input token id of each row of `documents.npy`
`queries.npy`	float32	`[2931, 32, 128]`	Query vectors
`queries_ids.npy`	int64	`[2931]`	LoTTE qid of each query (0 to 2930)

Document ids are LoTTE passage ids and equal the row index: row i of doclens.npy is passage i.

documents.npy stores float16 values as their raw 16-bit patterns, with dtype uint16. In numpy, read it with np.load("documents.npy", mmap_mode="r").view(np.float16).

Relevance judgments

The official LoTTE pooled dev search qrels: 8,573 relevant query-passage pairs over the 2,931 queries, all with relevance 1 (ir_datasets lotte/pooled/dev/search). Standard metric: Success@5.

Downloads last month: 62

Collection including tuskanny/lotte_pooled_colbertv2

multivector_datasets

Collection

54 items • Updated 9 days ago