Datasets:
Tasks:
Text Retrieval
Modalities:
Text
Formats:
json
Sub-tasks:
document-retrieval
Size:
100K - 1M
ArXiv:
Tags:
information-retrieval
hard-negatives
knowledge-distillation
sparse-retrieval
multilingual
miracl
License:
Commit History
2 days ago
-
Card: first/second set instead of round 1/2; runnable rebuild of the training file; plainer build steps
Verified signature732db48 -
Card: shorter; 56,195 distinct questions behind the 88,165 rows; teacher read at most 192 tokens per pair; Mr.TyDi BM25 negatives kept in sampled pools
Verified signature7ef0925
12 days ago
-
Card: absolute LICENSE links (relative links to extensionless files 404 on the Hub)
Verified signature9d836ad
13 days ago
-
Initial upload
Verified signature7e8154b -
initial commit
Verified signaturefc54452