All of our datasets (released and evaluation dataset) are sampled from the same billion-scale vector dataset, which consists of Bing queries encoded by Turing AGI v5 that trains Transformers to capture similarity of intent in web search queries.

#	Name	Description	Size
1	dummy-data.bin	dummy data for packing submission in reprozip	104
2	contest-data-release-1m.bin	medium scale released data	106
3	contest-data-release-10m.bin	large scale released data	107
4	secret-1m.bin	medium scale data, used for evaluation before March 10	106
5	secret-10m.bin	large scale data, used for evaluation after March 10	107

You can use AzCopy for downloading large scale datasets.