Datasets:
Tasks:
Summarization
Multilinguality:
multilingual
Size Categories:
unknown
Language Creators:
unknown
Annotations Creators:
none
Source Datasets:
original
Tags:
License:
sampled_validation / test
#2
by
Muennighoff
- opened
What's the purpose of these splits?
The purpose is that the original dev/test sets are massive and computationally expensive to run on. With the smaller sets you will still get statistically significant system comparisons, but for a fraction of the cost.
Muennighoff
changed discussion status to
closed